Estimating the costs of nuclear power: benchmarks and
uncertainties
François Lévêque
To cite this version:
François Lévêque. Estimating the costs of nuclear power: benchmarks and uncertainties. 2013.
<hal-00782190v2>
HAL Id: hal-00782190
https://hal-mines-paristech.archives-ouvertes.fr/hal-00782190v2
Submitted on 27 May 2013
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Interdisciplinary Institute for Innovation
Estimating the costs of nuclear
power: benchmarks and
uncertainties
François Lévêque
Working Paper 13-ME-01
May 2013
CERNA, MINES ParisTech
60 boulevard Saint Michel
75006 Paris, France
Email: [email protected]
Estimating the costs of nuclear power: benchmarks and
uncertainties
François Lévêque
Introduction
The debate on this topic is fairly confusing. Some present electricity production using nuclear
power as an affordable solution, others maintain it is too expensive. These widely divergent
views prompt fears among consumers and voters that they are being manipulated: each side
is just defending its own interests and the true cost of nuclear power is being concealed.
Companies and non-government organizations certainly adopt whatever position suits them
best. But at the same time, the notion of just one ‘true’ cost is misleading. As we shall see in
this paper there is no such thing as the cost of nuclear power: we must reason in terms of
costs and draw a distinction between a private cost and a social cost. The private cost is
what an operator examines before deciding whether it is opportune to build a new nuclear
power station. This cost varies between different investors, particularly as a function of their
attitude to risks. On the other hand the social cost weighs on society, which may take into
account the risk of proliferation, or the benefits of avoiding carbon-dioxide emissions, among
others. The cost of actually building new plant differs from one country to the next. So
deciding whether nuclear power is profitable or not, a benefit for society or not, does not
involve determining the real cost, but rather compiling data, developing methods and
formulating hypotheses. It is not as easy as inundating the general public with contradictory
figures, but it is a more effective way of casting light on economic decisions by industry and
government.
Without evaluating the costs it is impossible to establish the cost price, required to compare
electricity production using nuclear power and rival technologies. Would it be preferable to
build a gas-powered plant, a nuclear reactor or a wind farm? Which technology yields the
lowest cost per KWh? Under what conditions – financial terms, regulatory framework, carbon
pricing – will private investors see an adequate return on nuclear power? In terms of the
general interest, how does taking account of the cost of decommissioning and storing waste
affect the competitiveness of nuclear power?
This paperanswers these questions in three stages. We shall start by taking a close look at
the various items of cost associated with nuclear power. We shall look at how sensitive they
are to various factors (among others the discount rate and price of fuel) in order to
understand the substantial variations they display. We shall then review changes in the cost
dynamic. From a historical perspective nuclear technology has been characterized by rising
costs and it seems most likely that this trend will continue, being largely related to concerns
about safety. Finally we shall analyse the poor cost-competitiveness of nuclear power, which
provides critics of this technology with a compelling argument.
Adding up costs
Is the cost per MWh generated by existing French nuclear power stations €32 or €49? Does
building a next-generation EPR reactor represent an investment of about €2,000 per kW, or
twice that amount?
1/36
The controversy about the cost price borne by EDF resurfaced when a new law on electricity
was passed in 20101, requiring France’s incumbent operator to sell part of the output from
its nuclear power plants to downstream competitors. Under this law the sale price is set by
the authorities and must reflect the production costs of existing facilities. GDF Suez, EDF’s
main competitor, put these costs at about €32 per MWh, whereas the operator reckoned its
costs were almost €20 higher. How can such a large difference be justified? Is it just a
matter of a buyer and a seller tossing numbers in the air, their sole concern being to
influence the government in order to obtain the most favourable terms? Or is one of the
figures right, the other wrong?
As for investments in new nuclear power plants, the figures are just as contradictory. Take
for example the European Pressurized Reactor, the third-generation reactor built by the
French company Areva. It was sold in Finland on the basis of a construction cost of €3 billion,
equivalent to about €2,000 per kW of installed capacity. Ultimately the real cost is likely to
be twice that amount. At Taishan, in China, where two EPRs are being built, the bill should
amount to about €4 billion, or roughly €2,400 per kW of installed capacity. How is it possible
for the cost of building the same plant to vary such much, simply by changing its
geographical location or timeframe?
The notion of cost
The disparity between these figures upsets the idea, firmly rooted in our minds, that cost
corresponds to a single, somehow objective value. Surely if one asks an economist to value a
good, he or she will pinpoint its cost like any good land surveyor. Unfortunately it does not
work like that. Unlike physical magnitudes, cost is not an objective given. It is not a distance
which can be assessed with a certain margin of error due to the poor accuracy of measuring
instruments, however sophisticated they may be; nor is it comparable to the invariant and
intrinsic mass of a body. Cost is more like weight. Any object, subject to the force of gravity,
will weigh less at a certain elevation than at sea level, and more at either Pole than at the
Equator. In the same way cost depends on where you stand. It will differ depending on
whether you adopt the position of a private investor or a public authority, on whether the
operator is subject to local competition or enjoys a monopoly; again it will vary depending on
a given country’s hydrocarbon resources, and so on. Change the frame of reference and the
cost will vary.
In economics opportunity plays the same role as gravity in physics. Faced with two mutually
exclusive options, an economic agent loses the opportunity to carry out one if he or she
chooses the other. If I go to the movies this evening I shall miss a concert or dinner with
friends. The cost of forgoing one of the options is known as the opportunity cost. As
economic agents must generally cope with non-binary options, the opportunity cost refers
more exactly to the value of the next-best (second-best) option forgone. As preferences are
variable (Peter would rather see a movie than spend the evening with friends; for John it is
the opposite), the opportunity cost depends on which economic agent is being considered. As
a result it is eminently variable. Ultimately there may be as many costs are there are
consumers or producers. Regarding our present concern, the cost of building nuclear power
plants in Russia, which exports gas, will be different from the cost borne by another state.
Investing in nuclear plants to generate electricity, rather than combined-cycle gas turbines,
enables locally produced gas to be directed to a more profitable outlet. The economic
concept of opportunity cost puts an end, once and for all, to any idea that cost might be an
objective, invariant magnitude.
Moreover, it should be borne in mind that cost relates, not to a good or service, but to a
decision or action. The opportunity cost is not the cost of something, rather the cost of doing
something. This of course applies to the cost of production, which is defined by economists
using an equation, the production function. This function expresses the relation, for a
1
French law dated 7 December 2010 on re-organization of the electricity market.
2/36
particular technology, between the quantity produced – a kWh for instance – and the
minimum production factors required to achieve such output: labour, capital, natural
resources. The production function enables us to determine the cost of an additional unit of
the good, or the marginal cost- the opportunity of this additional production being measured
against the decision not to produce-. The production function also allows us to determine the
fixed cost of production, this time compared with the alternative option of producing nothing
at all. Its cost is not zero, because before producing the first unit, it was necessary to invest
in buildings and machines. So, even if the infrastructure is not used, it must be paid for.
To assess officially the cost of a good or service, it is advisable to ask an accountant, using
the appropriate methods. An accountant will calculate direct costs, in other words the costs
directly related to the product (steel purchases in car manufacturing) and indirect costs (R&D
expenditure, overheads) depending on the prevailing rules on cost allocation. Accountants
will distinguish between operating and maintenance costs, capital expenditure drawing on
shareholders’ equity or on borrowing in order to make investments. For 2010 France’s Court
of Auditors2 estimated that the accounting cost, not including decommissioning, of electricity
production by EDF’s nuclear fleet amounted to €32.30 per MWh. This figure corresponds to
annual operating and maintenance expenditure of nearly €12 billion, to produce 408 TWh,
and €1.3 billion annual capital costs, restricted to provision for depreciation. Obviously the
production cost found by an accountant depends on the method used. Using the full cost
accounting method for production the Court of Auditors found a total cost of €39.80 per
MWh. This figure is higher than the previous one, because the first method, cited above, only
includes depreciation in the capital costs, but does not allow for the fact that the fleet would
cost more, in constant euros, to build now than it did in the past. With the full cost
accounting method for production, assets not yet depreciated are remunerated and the initial
investment is paid back in constant currency.
In the another paper we shall take a detailed look at calculating the cost per kWh of
generating nuclear electricity with France’s existing capacity. For the time being we may
simply observe that neither an approach based on accountancy nor on economics yields a
single cost. For one kWh of nuclear electricity, much as for any other good or action, the idea
of a true or intrinsic cost for which accountants or economists can suggest an approximate
solution is misleading. On the other hand, as we shall see, their methods do help to
understand variations in costs, identify the factors which determine costs, compare such
costs for different technologies, and also observe the efficiency of operators. All these data
are valuable, indeed necessary to decide whether or not to invest in one or other electricity
generating technology.
Social, external and private costs
So cost is not invariant. Sometimes it is quite simply impossible to put a figure on it. This
additional complication concerns the external effects of using nuclear power generation, be
they negative – such as the unavoidable production of radioactive waste and damage in the
event of accident – or positive – avoiding CO2 emissions and reducing energy dependency.
Such external effects (or externalities in the jargon of economics) explain the disparity
between the private cost, borne by producers or consumers, and the social cost, borne by
society as a whole.
Economic theory requires us to fill the gap. Because of this disparity, the decisions taken by
households and businesses are no longer optimal in terms of the general interest; their
decisions no longer maximize wealth for the whole of society. For example, if it costs €10
less per MWh to generate electricity using coal rather than gas, but the cost of the damage
caused by emissions from coal is €11 higher, it would be better to replace coal with gas.
Otherwise society loses €1 for every MWh generated. But in the absence of a tax or some
2
Cour des Comptes, Les Coûts de la Filière Nucléaire, topical public report, January 2012.
3/36
other instrument charging for carbon emissions, private investors will opt to build coal-fired
power stations. Hence the economic precept of internalizing external effects.
How, then, are externalities to be valued in order to determine the social cost of nuclear
power? How much does it cost to decommission reactors and store long-term waste? What
price should be set for releasing one tonne of carbon into the atmosphere? How can the cost
of a major nuclear accident be estimated? What method should be used to calculate the
external effects of nuclear power generation on security with respect to energy independence
or the risk of proliferation?
We shall see that the answers to these question raise not so much theoretical or conceptual
issues, as practical difficulties posed by the lack of data and information. As a result, the
positive and negative external effects of nuclear power are only partly internalized. But then
the same is true of other sources of energy.
External effects relating to independence and security
We shall start with the trickiest question: putting a figure on the effects of national
independence. This is such a complicated task that no one has ever attempted it. Analysis so
far has only been qualitative. We often hear that nuclear electricity production contributes to
the energy independence of the country developing it. It purportedly yields greater energy
security. Many political initiatives are justified by such allegations, but the terms of the
debate are muddled. Conventionally energy dependence refers to the supply of oil products.
The latter weigh down the balance of trade of importing countries and subject them to price
shocks and the risk of shortages in the event of international conflict. Nuclear electricity
production only replaces oil and its derivatives in a marginal way. Only 5% of the electricity
generated worldwide is produced using oil derivatives.
In fact it would make more sense to look at gas, in order to justify the claim that nuclear
power contributes to energy independence and security. In this respect Europe, for example,
is dependent on a small number of exporting countries. The European Union imports twothirds of the gas it requires and the Russian Federation is its main supplier. Everyone
remembers the disruption of Russian gas transit through Ukraine in the winter of 2008-9. As
a knock-on effect gas deliveries in Europe were held up for almost three weeks. Millions of
Poles, Hungarians and Bulgarians were deprived of heating and hundreds of factories ground
to a halt. There is no doubt that Poland’s determination sooner or later to start nuclear
electricity production is partly due to the need to reduce its dependence on Russia. On the
other hand we have heard no mention of calculations putting a figure on the expected
benefit: a calculation resulting in acceptance, for instance, of nuclear power costing €5 or
€10 per MWh more than that of electricity generated using imported gas. The concepts of
energy independence and security are too fuzzy to measure. The best one can do is estimate
the cost of the shortfall for the Polish economy per day of disrupted supply. But to calculate
the gain in independence, this cost would have to be multiplied by the probability of such
disruption. However in 41 years Russia has only failed to honour its commitments twice, with
one interruption lasting two days, and the other 20. It would be difficult, on the basis of such
a small number of events, to extrapolate a probability for the future.
To take full account of security issues, allowance must be made for the risk of military or
terrorist attacks, and the risk of proliferation. In this case the externality is negative and
could counterbalance nuclear power’s advantage in terms of energy independence. A nuclear
power station is vulnerable to hostile action. For example, during the Iran-Iraq war in 198088 the nuclear plant being built at Busheir in Iran was bombed several times by Iraqi forces.
All other things being equal, the higher the number of nuclear plants in a country, the larger
the number of targets available to enemy action.
The development of civilian applications for the atom may entail the additional risk of
facilitating proliferation of nuclear weapons. Nuclear weapons can be manufactured using
plutonium or highly enriched uranium. The latter may be obtained by using and stepping up
enrichment capacity that already exists for producing fuel for nuclear reactors. Such fuel
4/36
must contain about 5% of uranium-235, whereas the concentration of this fissile isotope
must exceed 80% in order to produce a bomb. Plutonium is obtained from reprocessing
spent fuel.
No country has so far used fissile material from commercial reactors to produce weapons.
Reactors used – purportedly at least – for civilian research have however been used to
produce plutonium which can be used in weapons. India and North Korea are two instances
of this diversion. Iran’s nuclear programme also substantiates the claim that civilian nuclear
materials may be diverted toward military purposes. According to many observers the
development of commercial nuclear power is a cover for the production of fissile material to
make weapons.
One way of reducing the risk of proliferation would be to guarantee countries launching
programmes to develop nuclear power a supply of fuel for their reactors. In this way they
would no longer need their own enrichment capacity. This measure would restrict the spread
of enrichment technology, which can be diverted from its original civilian purpose. The
United Arab Emirates has, for example, made a commitment not to produce its own fuel. The
UAE will import it from South Korea, which is supplying turnkey reactors. Similarly the
Russians will guarantee a supply of fuel for the nuclear plants they are due to build in
Turkey. However agreements of this sort are contrary to the goal of reducing energy
dependence often associated with the decision to resort to nuclear power, there being only a
limited number of potential fuel suppliers. In the long run some countries will want to have
their own enrichment units, at least once they have a sufficient number of reactors.
There are no firm figures for the external effects of nuclear power on national and
international security, no more than there are for energy independence. Any attempt to
calculate such figures is thwarted by the scope of these concepts, both too broad and too
fuzzy. It would probably be wiser to leave it up to the diplomats and military strategists to
persuade their governments – using qualitative arguments – to revise, upwards or
downwards, the cost of resorting to nuclear power in their country.
The price of carbon
How are we to assess nuclear power’s contribution to combating global warming? Stated in
these terms the question is too general to allow an economist to provide an accurate answer.
There are too many uncertainties regarding the goal being sought and the consequences if
no action is taken. How is global warming to be defined? How large a share should be
attributed to human activity? Which greenhouse gases should be taken into account? We are
to back to the previous problem. On the other hand, values may be suggested for the benefit
of nuclear power in relation to reducing CO2 emissions to the atmosphere. To calculate this
benefit we need to know the price per (metric) tonne of carbon emissions, which can then be
combined with the emissions avoided for each MWh generated. At first sight this seems
straightforward. At a theoretical level, all the textbooks on environmental economics explain
how to determine the optimal price of a pollutant. In practical terms trade in CO2 emissions
credits provides an indication of the price of carbon. But in fact, the problem is still a thorny
one: we lack the data to apply the theory and the carbon markets produce the wrong price
signals.
The theory for determining the optimal price of a pollutant emission is simple enough in
principle. The optimal price is found at the point where the curve plotting the marginal cost
of pollution abatement intersects the curve for the marginal benefit of the avoided damage.
The general idea is that the level of pollution which is economically satisfactory for society is
the point beyond which further abatement costs more than the benefit from avoiding
additional damage. Or put the other way round, it is the point below which the situation
would not serve the public interest, the cost of additional abatement being lower than the
benefit it would yield; abatement is consequently worth carrying out. This coincides with a
basic economic principle according to which all actions for which the social cost is lower than
the social benefit should be carried out. As is the case with any equilibrium, the optimal price
5/36
corresponds to the optimal amount of pollution. Normative economics does not prescribe
zero pollution. The economically optimal amount of waste or effluents is only equal to zero in
the rare event of it being less expensive to eliminate pollution, down to the last gram, rather
than suffering the damage it entails.
Applying this theory is another matter. The data required to plot curves for CO2 emissions
abatement and avoided damages does not exist. Obviously there are estimates of the cost of
various actions such as insulating homes or recycling waste, which in turn limit carbon
emissions to the atmosphere. But economists need future costs, not just current costs. The
former are unknown, because technological innovation – such as carbon capture and storage
– has yet to yield conclusive results. It would also be necessary to know the cost of
measures to adapt to global warming. It may be more economical, at least for part of the
temperature increase, to adapt to the situation rather than combating it. But it is future
generations which will have to adapt. How can we know how much it will cost them? We
cannot ask them. The same applies to the damages suffered by our descendants. How could
they be calculated without an exact idea of their extent and without questioning those who
might be exposed to them? For example the cost of migration to escape changing
geographical conditions depends on the individuals concerned, in particular how much value
they attach to the loss of their land. The last, but no means the smallest, obstacle to
assessing damages is the lack of a robust formula for converting the concentration of CO2 in
the atmosphere into temperature increase. It is not the amount of carbon which causes the
economic loss but the climate change it may bring about. In this situation economists are
dependent on the scientific knowledge of climatologists. Unfortunately analysis of the exact
consequences for climate change of a rise in the amount of greenhouse gas stored in the
atmosphere is still tentative.
But does looking at the markets makes it any easier to find the price of carbon?
At first sight, yes. Since 2005 Europe has had a market for tradable emissions permits. On
this market the price of a tonne of CO2 fluctuated on either side of €15 in 2009-10
(equivalent to €55 a tonne of carbon, a tonne of CO2 containing 272 kilograms of this
element). Given that generating one MWh using coal releases roughly a tonne of CO2, the
operators of coal-fired plants had to pay an average of €11 per MWh for their emissions in
2009-10. In other words, all other things being equal, if in the course of this period, 1 MWh
generated by a coal-fired power station had been replaced by 1 MWh generated by a nuclear
plant, €11 would have been saved. Projecting ourselves into the future and anticipating that
the market price of carbon will double, switching electricity production from coal-fired to
nuclear power stations would save €22 per MWh.
So far, so good. On the basis of the market price we can obtain the opportunity cost we
sought, be it past or future. In addition the private and social costs seem to have been
reconciled: private operators are forced to make allowance for the price of carbon emissions
when choosing to invest in coal-fired or nuclear plants. Thanks to the market, the external
effect has been internalized.
In fact, nothing has been settled. For two reasons. Firstly, the European Emissions Trading
Scheme is not a market for polluters and polluteds, but an exchange for companies at the
source of emissions. It reflects the abatement performance of the various players, but in no
way the damage done. Secondly the market was badly calibrated. The prices it reveals are
not sufficient to achieve the targets set by the EU for reducing CO2 emissions. We shall now
take a closer look at these two reasons.
Economic theory explains that externalities occur in the absence of a market, so the answer
is to design one. With no market, there is no price, hence no purchasing cost and no
accountable expenditure. When manufacturers discharge harmful emissions into the
atmosphere, they are using the latter as a huge tip, access to which is free of charge. Some
polluters are not against the principle of a toll system, particularly for the sake of their
image. Similarly, to improve the market value of their home, some residents would be
prepared to pay polluters to restrict their emissions. But there being no marketplace where
6/36
polluters and polluteds can meet, pollution is free; it appears in no accounting system and
remains an external cost.
The European market for tradable emissions permits is not a place of exchange between
polluters and polluted. On the contrary it brings together companies to which individual CO2
emissions quotas have been distributed, but for a total amount capped below the level of
industry’s overall emissions. Let us suppose that, for example, 100,000 permits, each for a
tonne of CO2, are allocated whereas emissions from polluting companies amount to 120,000
tonnes. In this fictitious case, the companies would have to reduce emissions by 20,000
tonnes. In some companies the cost of cutting CO2 emissions is low, for others it is higher.
The first group will become sellers and carry out more abatement, the second group will buy
permits and abate less. At equilibrium, the price will be equal to the marginal cost of
eliminating the last tonne required to meet the limit. The advantage of this market is that
expenditure by industry on cutting emissions is minimized. Economic theory demonstrates
that a tax yields a similar benefit. With a tax on each unit of pollution, companies with low
costs for cutting emissions will abate more to reduce their liability for taxation; on the other
hand companies with high abatement costs will pay proportionately more tax and do less to
cut emissions. The main difference between a permit and a tax is the initial variable selected
by the competent authority. In the case of a tax, the price is set in advance and deploying
the instrument will show ex post the corresponding cut in emissions. For example a tax
pegged at €20 a tonne will lead to a 20,000 tonnes drop in emissions. For a permit, the
amount is decided first and the market then reveals the price per tonne of emissions
avoided. If an upper limit of 100,000 tonnes is set to reduce emissions by 20,000 tonnes the
market will balance out at a price of €20 a tonne.
The decision to base the system on price or quantity is closely related to the political
consensus underpinning the action. In the first instance agreement was reached on the level
of the acceptable surcharge per unit, in particular for consumers and business. Here the
unknown factor was the amount of abatement; it might be too low, but in any case the
economic conditions were such that a higher surcharge could not be applied. In the second
instance agreement was reached on the level of a significant reduction that needed to be
achieved, in particular according to scientific experts. The unknown was the price to be paid
for such a reduction, but in any case setting a lower target for pollution abatement would
certainly not have achieved the desired environmental effect. This is obviously an
oversimplification. In the absence of accurate data on the cost of abatement, orders of
magnitude may sometimes be posited. When the initial level of taxation is announced,
business and government may be able to estimate how much pollution will be abated within
a certain range. In the other case, when the initial abatement target is published, the various
players can estimate an approximate price. Once the first variable has been set, the second
is not usually completely unpredictable.
However, as the European example (see box) shows, the initial calibration may be faulty.
The failures of the EU Emissions Trading Scheme
By mid 2013 the price of CO2 had dropped to less than €5 a tonne. Five years ago the
European Commission predicted that by this point in time it would be worth €30. The
financial crisis and the drop in industrial output obviously explain part of the difference. But
in 2006 the price had already fallen below €15 a tonne. The main structural reason behind
the persistently low price of CO2 is the failure to create scarcity, too many permits having
been distributed. The resulting downward pressure on prices has been exacerbated by lower
than expected emission-abatement costs. The European CO2 trading system does not fulfil its
purpose: it does not send a reliable signal enabling industry to curtail long-term investments,
in particular enabling electricity utilities to choose between various generating technologies
according to their CO2 emissions performance.
The case of the United Kingdom is a perfect illustration of this failure. The UK has
undertaken to halve CO2 emissions by 2030. To achieve this target it plans to set a carbon-
7/36
price floor at €20 a tonne in April 2013, slated to double by 2020, ultimately reaching €87 a
tonne by 2030. The price in the EU Emissions Trading Scheme will only exert an influence if
it exceeds these price-floors, which is unlikely to happen very often unless the ETS is
reformed in the meantime. France offers another example. In 2009 the government was
planning to introduce a carbon tax as an incentive to reduce the use of oil products. The
rates recommended to achieve a fourfold cut in emissions by 2050 were €32 a tonne in
2010, rising to €100 a tonne in 2030, and twice that amount by mid-century3. The ETS price
for carbon is far below the value recommended by experts to achieve long-term targets for
reducing emissions.
The preceding discussion of the price of carbon is important, as we shall see, for it is one of
the determining factors in the competitiveness of nuclear power: without taxes on CO2
emissions or in the absence of an emissions trading scheme, nuclear power cannot compete
with coal or even gas. Furthermore, a consideration of how the cost of carbon is assessed
highlights the dual role played by economic analysis. In the world of perfect information
posited by economic theory, such analysis would enable us to set the optimal level of
abatement, at the intersection between the cost of the damage done by an additional tonne
of emissions and the cost of reducing pollution by an additional tonne. The role of
government would be simply to plot curves and enforce the resulting target price or quantity.
The economic analysis would dictate its prescriptions to policy-makers. In the real world of
limited information, in which we live, economics occupies a humbler position and the roles
are reversed. Political decisions, through voting, debate or consultation lead to the definition
of an acceptable level of either damages or expenditure. Economic analysis only intervenes
to minimize the cost of achieving the degree of damage decided by government or to
maximize the quantity produced corresponding to the level of expenditure set by
government.
Decommissioning and waste: setting the right discount rate
Nuclear electricity generators are responsible for the waste and by-products they produce. In
this field, much as elsewhere, the polluter-pays principle applies. Nor is this principle
disputed by the operators of nuclear plants, nor yet by opponents of nuclear power. So the
controversy does not centre on the need to internalize the costs of decommissioning reactors
and storing waste (spent fuel, decommissioning debris), but on the amount to be set aside
now to cover these costs, in order to ensure that these back-end activities can be carried out
tomorrow.
Worldwide we have almost no experience of dismantling power stations and burying
radioactive waste. Nowhere in the world has anyone so far built a permanent storage facility
for burying long-term waste. In France not a single nuclear power station has been
completely decommissioned. Work decommissioning the Chooz A reactor, in the Ardennes, is
only scheduled to end in 2019. The reactor was commissioned in 1967 and shut down 24
years later. Worldwide less than 20 commercial reactors have been completely dismantled.
The lack of references makes appraisal very uncertain. We cannot rule out the possibility
that the technical costs of dismantling and the costs of waste-management may prove very
high. However, even if this were the case, it would have little effect on the return on
investment from a new nuclear power station. The return is not very sensitive to this
parameter because the costs at the end of a nuclear plant’s service life are very remote in
time, and a euro tomorrow is worth less than a euro today, and even less the day after
tomorrow. Future costs or benefits are wiped out by the rate of exchange used to convert
present funds into future funds (or vice versa). For example, at an annual rate of 8%, €1
million would only be worth €455 in a century. This amount drops to €0.20 after two
3
Centre d’Analyse Stratégique, La Valeur Tutélaire du Carbone. Report by the committee chaired by
Alain Quinet, La Documentation Française, n° 16, 2009.
8/36
centuries and in 500 years it would have dwindled to almost nothing. If the plant had to be
decommissioned now, taking the same rate and supposing that decommissioning would cost
15% of the total cost of a new reactor, this share would only represent 0.7% of the total cost
if work was carried out 40 years later. This rate, known as the discount rate, plays a decisive
part in assessing the costs of decommissioning plant and managing waste. To avoid wiping
out such costs, a discount rate close to zero would need to be used. Certain environmental
conservation groups advocate this position, but there is little support among economists. We
shall now look in greater detail at how the discount rate works.
To avoid confusion, we should start by explaining what this rate is not. Firstly the discount
rate bears no relation to inflation. The latter, whether its origin is monetary or results from
indexing wages, is a phenomenon which raises prices. Consumers will buy less tomorrow
because the same shopping basket will cost more. Secondly the discount rate does not
reflect the risks associated with the investment project being assessed. Such risks cast doubt
on income and expenditure and change the way they are estimated, but not due to the
discount rate.
To convert current euros into future euros we must start from existing knowledge. Despite
the limited experience mentioned above, we do have preliminary orders of magnitude. In
2010 France’s Court of Auditors used the EDF estimate of how much it would cost to
decommission its 58 reactors. The cost entered in the company’s accounts amounts to €18
billion, equivalent to €300 per kW of installed capacity. In comparison with assessments in
other countries, and consequently relating to reactors and conditions which may be very
different, this figure is near the lower end of the range. The management consultants Arthur
D. Little estimated that the upper value in Germany would be close to €1,000 per kW. In the
United States estimates of the cost of decommissioning the Maine Yankee plant, completed
in 2005, are in the region of €500 per kW. As for waste destined to be buried in deep
geological repositories, only very preliminary estimates have been made. Work is still
focusing on pilot schemes or has barely started. The only site currently operating stores
radioactive waste of military origin at Carlsbad, New Mexico. This waste is easier to manage
because it does not release any heat. To store the amount of long-term waste produced by a
reactor in one year the order of magnitude currently cited is €20 million. This figure is based
on various British, Japanese and French estimates4. The amount is likely to change with
progress by research and technical know-how. In 2005 France’s Nuclear Waste Authority
(Andra) estimated that it would cost a little under €20 billion to build and operate a deep
geological repository. Five years later adjusted new assessment was made up to €35 billion.
The second amount makes allowance for additional parameters, integrating return-onexperience from excavating underground galleries, requirements for greater capacity and
tougher safety constraints, among others.
The timescales we are dealing with here are very long. Some categories of nuclear waste will
go on emitting radiation for several hundreds of thousands of years. For example plutonium239 has a half-life – the time required for half the radioactive atoms to disintegrate – of
24,000 years. For technetium-99 it rises to 211,000 years and for iodine-129 it is 15,7
million. Such time spans are stupendous when compared to the scale of human life. Our
most distant ancestors, Australopithecus, appeared on Earth 4 million years ago and modern
humans (homo sapiens) only emerged about 200,000 years ago. Of course there are no
plans for the storage facilities to operate for such long periods. For example the deep
geological repository projected by Andra is expected to last for 120 years, from the start of
construction to final closure. If a decision was taken now to invest in a new reactor, the plant
would be commissioned in 2020 and operate for 60 years. Only in 2100 would
decommissioning be complete, with the last tonnes of waste finally being buried in 2200.
These economic deadlines are short compared with the half-life of certain waste products,
but nevertheless dizzying. A century is a very long time, in the life of an economy, with its
4
Cour des Comptes, op cit, p150.
9/36
multiple crises. Government bonds, the investments with the longest time span, spread over
periods of 20 or sometimes 30 years, stretching to 50 in exceptional cases.
With such long timeframes, how can we account for this expenditure now? Utilities must
make provision for such liability in their accounts, integrating the cost in calculations of the
social rate of return on generating nuclear electricity.
The reference to government bonds suggests a preliminary approach to the discount rate
and its basis. If someone offers to give you €100 today or in 20 years time, there is no need
to think twice. If you take the €100 now you can make a very sound investment in US
Treasury bonds. Thanks to the interest you will have more than €100 in 20 years. You will
thus be able to consume more than if you had agreed to wait before receiving the funds. The
decision, based on a simple trade-off, justifies the use of discounting and the long-term
interest rate may be used to find the future value of today’s euros. With 4% interest, €100
today will be worth €220 in 20 years or, inversely, €100 from 20 years ahead would be
worth only €45.60 at present. However using this interest rate to discount the value does not
solve the problem, if the aim is to determine the value of a euro in a century. As the
business weekly The Economist amusingly observed5, “At a modest 2% rate [...] a single
cent rendered unto Caesar in Jesus’ time is equivalent to [...] 30 times the value of the
entire world economy today”.
Furthermore interest rates only partly justify discounting. According to economic theory
discounting is necessary for two reasons: people are impatient and future generations will be
better off. The economic agents featuring in the models display a pure time preference for
the present. Instead of taking an interchangeable value, let us suppose that the choice
concerns the possibility of attending the performance of an opera in the course of the coming
year, or the same performance in five years’ time. Which ticket would you choose? The
interest rate argument does not hold because you can neither loan nor resell the ticket. If
you do not use it, it will be wasted. It is highly likely that you will opt for the performance in
the coming year, rather than waiting five years. This impatience is reflected in a pure time
preference for the present, which crushes future consumption to give it less weight.
It is more difficult to illustrate the notion that future generations will be better off. The
discount rate depends on the growth rate of the economy and a barbaric term, the elasticity
of inter-temporal substitution in consumption. The overall idea is that the richer you are, the
less satisfaction an additional euro will yield. If you give €100 to someone with a low income,
you will be making him a present worth much more than if you give the same amount to a
millionaire. Marginal utility decreases with income. Consequently if society is €1 billion richer
tomorrow, it will respond less to this gain than now. With an ordinary utility function, society
10 times better off than at present, and elasticity equal to 1, contemporary society would
see its well-being increase 10 times more for each marginal unit of consumption (€1 billion)
than tomorrow’s society. So it would be advisable to limit our efforts to provide benefits for
future generations.
Economists thus provide the following key, forged by Ramsey in 19286, for calculating the
discount rate: the discount rate (d) is equal to the sum of the pure preference rate for the
present (p) and the product of the elasticity of the marginal utility due to consumption (e)
multiplied by the growth rate of per capita GDP (g), in other words d=p+eg. With the three
values often used [2,2,2], the discount rate is 6%.
Interpretation of the three variables merits closer attention.
The pure preference rate for the present may be seen as an equity parameter, its value
depending on how fairly we wish to treat future generations. Let us suppose that the output
from a new nuclear plant entails a waste-management cost of 100 in a century, but yields a
present gain because nuclear technology is cheaper. If we want to treat future generations
even-handedly, we should only commit ourselves to the investment if its present benefit for
5
The Economist, Is it worth it, 3 December 2009
6
F.P. Ramsey, 1928, A Mathematical Theory of Saving, Economic Journal, 38, p543-559.
10/36
us is greater than 100. In this case we would apply a pure preference rate equal to zero.
This position in favour of equality between generations is defended by some economists,
including Ramsey. It rejects the idea of discrimination depending on the date of birth and
involves treating all generations on the same footing, even if they are more prosperous. If
we want to treat our descendants slightly less favourably (if, for example, we are convinced
they will find smarter means of storage or recycling), we need to use a slightly positive
preference rate. A gain of 14 today will suffice (equal to 100 discounted at 2% over 100
years). If on the other hand we are feeling selfish and have no concern for what comes after,
the rate will be very high: even if nuclear power only yields today a unit gain, it is worth
taking, its value exceeding 100 in the future (discounted at 8% a year, it will be worth €0.40
in a century). We may also interpret the pure preference rate in terms of our chances of
survival. With a one in ten chance of mankind not surviving for 100 years (following, for
example, collision with a meteorite), the value of the preference rate is 0.1; it rises to 1 if we
assume the likelihood of survival is 0.6 (a 4 in 10 chance of the end of the world).
The elasticity of the marginal utility due to consumption also measures equity. The greater
the difference in utility for a marginal unit of consumption between low and high-income
households, the more justification there is for high levels of transfer, through taxation for
instance, from rich to poor. Such transfers raise the utility of the whole of society. In other
words, this parameter reflects our attitude to unequal levels of consumption, between
different people in the present day, or between them and their descendants. Unlike the
previous variable, the difference in treatment is not related to time. The more egalitarian we
are the more we favour redistribution from rich to poor and the higher the value we need to
use for elasticity when calculating the discount rate. If we assume that future generations
will be richer than today, it is legitimate to limit our efforts to improve their welfare. On the
other hand, elasticity equal to 1 is unfair. Given a constant population it would justify
spending 1% of today’s GDP to give future generations the benefit of an additional 1% of
GDP, even if they are incomparably more prosperous. Per capita fractions of GDP can
therefore be traded between generations on equal terms. The elasticity of marginal utility
may also be linked to risk. According to economic theory risk aversion is proportional to
elasticity. The higher the elasticity, the more a person is prepared to pay for the certainty of
consuming 100, rather than a random outcome (for example, a one-in-two chance of
consuming 200, or zero). Taking a higher value for this parameter, which is then multiplied
by the growth rate of per capita GDP is tantamount to assuming that the present generation
is averse to risk.
Setting the three values which make up the discount rate is no easy matter; but they will
play a decisive role in how we act now. An instance of this point is the controversy prompted
by the publication in 2006 of the Stern Review7. This report caused quite a stir because it
concluded that substantial, immediate expenditure (about 1% of GDP) was needed to reduce
greenhouse gas emissions. This recommendation contradicted the conclusions of most
climate-change economists which suggested a more gradual increase in expenditure. The
work of the US economist William D. Nordhaus8, for example, recommends a carbon tax of
$13 a tonne over an initial period in order to internalize the damage done by global warming.
Nicholas Stern prescribed $310 a tonne. Half of this difference is simply due to the discount
rate used by the two parties: 4% for the former, 1.4% for the latter.
In his estimate Stern uses a preference rate for the present of 0.1 and elasticity of 1. These
two values represent the lower limits of the ranges economists generally accept. His choice is
open to criticism because it raises a logical contradiction. A low preference rate for the
present should go hand-in-hand with high elasticity or, on the other hand, low elasticity
should match a high preference for the present. It would be mistaken to suppose that one of
these two parameters reflects equity between generations, the other solidarity within a
single generation. A low value for the elasticity of the marginal utility due to consumption
7
The Stern Review of the Economics of Climate Change, October 2006.
8
[note manquante ?]
11/36
can be justified on the grounds of reducing inequality between rich and poor, regardless of
when they were born. This choice coincides with a high preference rate for the present,
which endorses the idea that the present generation should only make limited sacrifices for
future generations (given that the latter will be better off, as Stern posits with a positive
growth rate for per capita GDP). Using a simplified economic model the Cambridge
economist, Partha Dasgupta9, has demonstrated that the parameters used by Stern would
lead to inconceivably high saving ratios. With a preference rate for the present of 0.1,
elasticity of 1, and a world with neither technological progress nor population growth, we
should be investing 97.5% of our current output in boosting the standard of living of future
generations. The Stern Review asks whether it makes economic sense to spend 1% of
today’s GDP to prevent damage amounting to 5% of GDP in a century. The three values it
uses, [0.1; 1; 3], would lead to a discounted benefit five times greater than the cost. But if
we use [2; 2; 2] the discounted benefit would be 10 times smaller than the cost! In other
words, with a 1.4% discount rate it is entirely justifiable to spend 1% of GDP on reducing
greenhouse gas emissions, whereas with a 6% discount rate it would be quite out of the
question.
We have so far set aside the question of the third parameter, the future growth rate of per
capita GDP. Its value is just as uncertain as the others, but setting it does not raise equityrelated issues. Looking back in time, the annual growth rate of per capita GDP was 1.4% in
the UK from 1870 to 2000, and 1.9% in France. However these averages conceal significant
variations. In the UK the growth rate was 1% in 1870-1913, 0.9% in 1913-50, 2.4% in
1950-73, and 1.8% from 1973 to 2000. Over a very long period of time – 1500 to 1820 – it
is estimated to have been 0.6%. Which of these different rates should we use? The growth
rates for the next century or two may be very different. Nor can we rule out a negative
growth rate, though it is not very likely. However global warming in excess of 6°C in 200
years could have precisely that effect.
The discount rate cuts both ways, exerting a decisive influence on decisions regarding public
and private investment, but it is impaired by numerous unknowns. One recent attempt to
reduce this tension has involved using a rate that varies over time – rather than being
constant – declining as it advances into the future. The per capita growth rate can be used to
illustrate the intuition behind this idea: the more remote the future, the greater the
uncertainty regarding economic and technological progress; and consequently the greater
our caution regarding action that might jeopardize the well-being of future generations, the
lower the discount rate should be. The French economist Christian Gollier10 recommends
using a 5% annual discount rate for costs borne over the next 30 years, dropping to 2% for
subsequent costs. It is also possible to set the discount rate on a downward path, with either
several steps or a steady decline. In a report submitted to the British government in 200211,
Oxera Consulting Ltd suggested adopting a 3.5% rate from 0 to 30 years, 3% from 31 to 75
years, 2.5% from 76 to 125 years, and so on, with the rate ultimately bottoming out at 1%
after 300 years. In France the Lebègue report12, on a review of discount rates in public
investment, recommended a 4% rate for the first 30 years, then a rate that would steadily
decrease to reach 3% after 100 years, tending towards 2% for a time horizon of over 300
years.
A varying rate also seems to represent a compromise between two demands: on the one
hand taking account of our preference for the present and our contribution through technical
progress to the prosperity of future generations; and on the other hand allowing for the
9
P. Dasgupta, Comments on the Stern Review’s Economics of Climate Change, November 2006, memo.
10
C. Gollier, Discounting an Uncertain Future”, Journal of Public Economics, 85, 2002,
p149-166.
11
A Social Time Preference Rate for Use in Long-Term Discounting, Oxera, December 2002.
12
Report by the group of experts led by Daniel Lebègue, Révision du Taux d’Actualisation des
Investissements Publics, Commissariat Général du Plan, January 2005.
12/36
potentially very negative consequences of our action, or inaction, with regard to future
generations13.
It is obviously up to the relevant authorities to ensure that adequate provision is made for
the projected costs of decommissioning and waste management, in accordance with the
discount rate they have decided. In both the United States and France the government took
such measures long ago. Left to themselves utilities would stand to gain by underestimating
future expenditure on this work and by opting for high discount rates in order to minimize
projected costs. In the US and France – but also in many other countries – today’s
consumers are paying for tomorrow’s expenditure. There are no hidden costs for
decommissioning and waste which once internalized would make the cost of nuclear
electricity production prohibitive (see box).
Taking into account the costs of decommissioning and waste
In France regulation is based on special discounted provisions imposed on EDF. They appear
on its balance sheet and the utility is required to secure them with specific cover assets. The
law sets an upper limit for the discount rate pegged to 30-year government bonds, currently
close to 3%. With this rate EDF’s provisions for decommissioning and waste amount to €28
billion. They would increase by 21% with a 2% discount rate, adding 0.8% to the overall cost
of a MWh. Furthermore, if just the cost of decommissioning was to rise by 50% (amounting
to €30 billion as opposed to €20 billion), the cost of electricity would increase by 2.5%14. If
the cost of deep geological repositories was to double, it would result in a 1% increase. In
the US a special fund has been set up to cope with the future expense of deep repositories
for spent fuel. Utilities pay a fee into the fund equal to $1 per MWh they generate. The
Department of Energy checks at regular intervals that the fee is sufficient. For this purpose it
has developed about 30 cashflow models designed to balance out by 213315. These scenarios
depend on a large number of parameters, including the discount rate. The lowest rate
considered is 2.24% per year. Two out of the four scenarios based on this rate result in a
deficit, whereas the proportion is only one in four for the scenarios using a higher rate. The
figures above are valid for existing reactors in the US and France.
For new nuclear plants, the time horizon for expenditure would be longer, so
decommissioning and waste-management costs would have even less impact on the present
value of projects. A Massachusetts Institute of Technology study16 on the future of nuclear
power puts the overnight cost of building a reactor at $4,000 per kW, and the cost of
decommissioning it at $700 per kW, or 17.5%. Spreading decommissioning expenditure out
between the 41st and 110th year after the reactor is commissioned, and assuming a 6%
discount rate, would bring the present value of decommissioning down to $11 per kW. This
value would be five times higher ($52) if the rate was almost halved (3.5%). But as before,
this cost is negligible compared with construction costs. With the above discount rates, the
17.5% shrinks to 1.3% or 0.2%, respectively.
So back-end activities have no significant impact on the cost competitiveness of existing or
new nuclear power. Unless of course one adopts a very low, or even zero, discount rate for
very distant time horizons – as is the case in the Stern Review’s calculations for climate
change. In our opinion, this stance – which its advocates justify by the hazardous nature of
nuclear waste and its very long life – boils down to using inconsistent economic reasoning to
endorse a legitimate argument. In this paper we have not allowed for the possibility that
such waste might represent a risk for future generations. The only waste-related costs taken
13
The intuitive, common sense solution of a declining, variable discount rate now has a solid base in
economic theory, see the book by C. Gollier, Pricing the Planet’s Future, The Economics of Discounting in
an Uncertain World, Princeton University Press, 2013.
14
Cour des Comptes, op cit, p282.
15
DoE, Civilian Radioactive Waste Management Fee Adequacy Assessment Report, July 2008. RW-0593.
16
Update of the MIT 2003 Future of Nuclear Power, 2009.
13/36
into account are the preventive costs built into the quality of repositories and their
supervision. These costs vary depending on the safety standards set by government for
decommissioning and storage. To take a trivial example, the cost of a repository increases in
relation to the length and depth of its tunnels. On the other hand, this calculation makes no
allowance for the cost of possible accidents, despite the fact the risk does exist.
At Fukushima Daiichi there could have been a loss of water from the cooling ponds
containing spent fuel, or even their collapse, leading to massive radioactive emissions.
Securing future generations against a disaster of this sort poses the problem of assessing the
uncertain damages associated with events with a very low probability and a very high cost.
The release of radioactive substances into the atmosphere following the meltdown of a
reactor core raises the same question: how is one to estimate the costs without knowing how
the risks are distributed. We shall address this key question in a companion paper. The
discount rate is of only limited value for finding an answer. A wrong way out would be to give
the matter no further thought and select a very low, or zero, value to allow for the hazards
of wasteMaking allowance for a possible disaster caused by downstream activities may mean
opposing the construction of new nuclear reactors without it being necessary to hide behind
a very low or zero discount rate.
Liability in the event of accident
Another paper devoted to risks and regulation deal in detail with the cost of major accidents
and the legal framework for the civil liability of nuclear power. But we need to mention the
matter briefly here, many authors having suggested that estimates of the cost of nuclear
power fail to make allowance for the risk of disaster.
The operators of nuclear power stations are liable in the event of accident, but it is true that
in most cases an upper limit is placed on such liability. The amount of compensation they
must pay in the event of massive radioactive emissions is less than the value of the
damages. In France, for instance, the limit is €91.5 million. It will soon be raised to €700
million. Such caps on liability raise the question of whether the costs of major accidents are
sufficiently internalized. According to the opponents of nuclear power, limited liability is
equivalent to a hidden subsidy. After all a Swedish study17 estimated that the Chernobyl
disaster cost nearly $400 billion. There is no way of settling the matter without a detailed
review of the expected and observed frequencies of accidents and the uncertainty
surrounding the level of damages. Here we shall make with a much simplified examination of
the risk involved, in order to determine, within this framework, how much impact it has on
the full cost of existing nuclear plants and new reactors.
Risk is classically defined as the result of multiplying the probability of an accident by the
severity of the outcome. For the sake of argument, we shall take the highest values cited in
the literature for these parameters. We shall suppose that there is one chance in 100,000
that a disaster may occur during one year of a reactor’s service life, a probability 100 times
higher than the figure cited by Areva for the EPR. We shall then suppose that the massive
release of radioactivity causes damage to public health and the environment worth
€1 trillion, 10 times higher than the provisional estimates for Fukushima. So the risk is equal
to 0.00001 x 1,000,000,000,000, or €10 million a year. Supposing that the reactor’s annual
production amounts to 10 million MWh, the risk would be equivalent to €1 per MWh, or 40
times less than the cost estimated by the regulatory authorities for nuclear electricity
generated by EDF, or indeed between 50 and 100 times less than the estimates of the
average cost of new nuclear. This scratch calculation shows that, under much simplified
conditions – in particular no allowance for uncertainty or aversion to risk – internalizing the
full cost of an accident has only a very slight impact on the cost of nuclear electricity.
17
Economic losses estimated at $148 billion for Ukraine and $235 billion for Belarus. Figures cited by Z.
Javorowski, The Chernobyl Disaster and How it has been Understood, 2010.
14/36
We should however point out that, on the basis of these hypothetical data, the upper limits
on liability currently in force mean that only a relatively small share of costs is internalized.
If we take the case of the €91.5 million limit in France, it only amounts to 0.4% of the full
cost of an accident18. Raising the limit to €700 million would still leave 97% unaccounted for.
In other words internalization is indeed partial, but internalizing the full cost would only
result in a slight increase in the cost of nuclear electricity.
Technical and financial production costs
Here at last we may venture onto more solid ground. Engineering economists do not base
their decisions on externalities which are so difficult to grasp and estimate. On the contrary
they work on data, relating in particular to the costs of reactors built in the past and current
operating costs. They can use proven, widely accepted methods for calculating costs, in
particular for project funding. They juggle with concrete, steel, enriched uranium, manmonths, assets and deadlines.
To come to grips with the subject we shall start with construction. This involves an overnight
cost and a capital cost. The overnight cost refers to a hypothetical construction project
completed in an instant, or ‘overnight’. Spending on material, machines and wages is
entered into the accounts at the prices in force when construction starts. This does not
overlook financial costs; they are simply processed separately. It takes from five to 10 years
to build a power station, from initial preparation of the site to the moment it is connected to
the grid. During this time there is no return on investment. On the contrary, it represents a
cost. If the operator borrows half the amount it needs from banks, at a 4% real interest rate
(allowing for inflation), and funds the rest out its own resources at 6%, for instance, the
average cost of capital is 5%. This cost must be added to the overnight cost to obtain the
cost of investment, or installed cost.
The overnight cost is useful if we want to make an abstraction of the variability of
construction lead-times. It makes comparisons easier, because construction times vary
depending on the reactor model and size, but also due to non-technical causes, particularly
changes in the prevailing regulatory framework or local opposition. In the US for example,
the shortest construction project lasted less than four years, but the longest one took 25
years.
Although it overlooks such factors the overnight cost can vary a great deal. Firstly, over
time. On a per kW basis the first reactors were much cheaper than at present. We shall
examine this dynamic in the following section. The overnight cost also varies in space. In its
2010 study of electricity production costs the OECD noted a difference of one to three
between the overnight costs, expressed in $ per MW, for building a reactor in South Korea
and Switzerland19. The size, model and country (cost of labour, regulatory framework, etc.)
are not the same, but such a large difference may nevertheless come as a surprise. However
it is not specific to nuclear power. The OECD observed a similar disparity for gas, with South
Korea and Switzerland once again at the two extremes20.
The overnight construction cost is one of the three main factors affecting the cost of
generating nuclear electricity. The other two are the load factor and the capital cost (see
box).
18
A fleet of 58 reactors with a 40-year service life, or an expected number of accidents of 2320 x 10-5 =
0.0232 and damages of 0.023 x €1,000 billion, in other words €23.2 billion.
19
Data compiled from 14 countries, of which three non-OECD, but not the US, p59 of the report.
20
Table 3C p61. CCGT.
15/36
Load factor and cost of capital
Nuclear power plants are characterized by very long construction times and a very high fixed
investment cost compared to a variable operating cost, particularly with respect to fuel
expenses.
As a result, if a reactor does not operate at full capacity, once it has been built, the fixed
cost must be paid off by a smaller amount of electricity production, which in turn means that
each MWh is more expensive. Over the past decade the load factor of existing nuclear plants
was about 95% in South Korea, 90% in the US and 70% in Japan. To illustrate the weight of
this factor, we may use an example from the book by Bertel and Naudet21: improving the
load factor from 75% to 85%, boosts output by 13% and cuts the cost of an MWh by 10%.
The cost of capital depends on how long it takes to build the plant, but also on the choice of
discount rate. As the overnight construction cost is spread over several years, expenditure
must be discounted. The calculation uses the date on which the plant was commissioned as
its baseline and a discount rate decided by the operator. The difference between this
discounted expenditure and the overnight cost is referred to as interim interest. It measures
the cost of capital. For a private-sector operator the discount rate may range from 5% to
12%. With construction lasting six years, the overnight cost must be multiplied by 1.16 with
a 5% discount rate, and by 1.31 with a 10% rate22. Obviously the sooner construction is
complete, the sooner income will start to flow in, with interim interest reduced accordingly.
In the example borrowed from Bertel and Naudet, shortening the construction time to five
years reduces the cost of capital by 27%, with a 10% discount rate, and by 13% with a 5%
discount.
Once construction of the plant is complete, expenditure concerns fuel and other operating
and maintenance costs. Roughly speaking fuel costs represent between 5% and 10% of the
cost of generating electricity, with the other costs totalling between 20% and 25%. The cost
of fuel varies depending on the amount of electricity generated, because it is depleted as the
chain reaction proceeds. It is this chain reaction which releases heat, used in turn to
generate electricity. The level of production has little impact on the other operating costs,
which may be treated as relatively fixed, at least as long as the reactor is in service. When
the nuclear plant is finally shut down, most of these costs disappear.
Adding up the costs: the levelized cost method
The technical and financial costs of building and operating a nuclear power plant, the
downstream costs of decommissioning and processing waste, and the external costs
(avoided carbon emissions, accidents) must all be added up to obtain the full cost of nuclear
power. It will then be possible to monitor variations in this cost over time and to compare it
with the cost of electricity generated using other technologies. To do so, we need to convert
the euros at different points in time into constant euros and MWs into MWhs. The discount
rate is used for the first conversion. The second operation is required in order to add up fixed
costs – expressed as value per unit of power, for example in € per MW – and variable costs –
expressed as value per unit of energy, for example in € per MWh. By definition, one MWh is
the amount of electricity generated by one MW of power in one hour. A 1,000 MW nuclear
plant operating at full capacity round the clock will generate 8,760,000 MWh a year. To
21
L'Economie du Nucléaire, by Evelyne Bertel and Gilbert Naudet, EDP Sciences, Paris, 2004, p57.
Obtained using Direction du Gaz, de l'Electricité et du Charbon (Digec) assumptions and an 8% discount
rate.
22
Bertel and Naudet, op cit, p116.
16/36
allocate investment costs we need to know or anticipate the plant’s load factor and its
projected service life.
The full cost is worth knowing, but what is really important is whether it is greater or less
than the revenues, in order to determine whether there is a net gain for the utility or any
other company venturing into nuclear power. So far we seem to have disregarded revenues.
Nor have we addressed the price of electricity and how it is sold. However, in conceptual
terms, there is no difference between a cost and a benefit. One switches back and forth
between them just by changing the sign. They are two sides of the same coin: a purchasing
cost for a producer is a source of income for its supplier; an avoided carbon emission cost is
a benefit for the environment.
Cost-benefit analysis, which compares discounted costs and benefits, is the canonical
method used by economists to estimate the private or social merits of a project or decision.
However a variant is used in the field of electricity, the levelized cost. It is used to determine
the price of electricity required to balance income and outgoings all through a power station’s
service life. In a way it takes the opposite route to the economics canon: instead of
calculating a project’s rate of return as a function of assumptions on the future price of
electricity, this variant sets a zero profit rate from which to deduce a price for electricity
which balances discounted income and outgoings. For example, taking €75 per MWh as the
levelized cost of the EPR plant at Flamanville, in western France, means it will break even if
the average price recorded reaches this level during the plant’s operational service life for
the projected number of hours’ operation. But bear in mind that zero profit does not mean
that there is no return on capital. The outgoings accounted for by this method include the
cost of bankers’ loans and raising funds from investors.
The levelized cost method goes back to before liberalization of the electricity sector and the
creation of wholesale electricity markets. It enabled a regulator to determine the sale price
of a monopolistic operator on the basis of the latter’s costs. It also allowed the two parties to
identify, by comparison, the cheapest generating technology in which to invest in order to
meet rising demand. For the economics of today’s electricity markets, only the comparison is
of any interest. In principle private operators, not government, take decisions on investment.
Operators tend to base such decisions on forecasts of future electricity prices and
consequently on the cost-benefit analysis. On the other hand, to decide whether it is
preferable to add coal or gas-fired, or nuclear plant to existing capacity, they will use the
levelized cost variant, because it makes it easier to compare technologies. In practice, even
after liberalization of the electricity market, government has continued to have a say in the
choice of generating technology. At the very least it plays a part in setting long-term targets
for decarbonizing electricity generation in line with policy on emissions abatement. In this
case the average discounted social cost will be used. Technical and financial costs, including
back-end costs (site remediation and waste management), are added to estimates of
external effects (such as accidents, pollutant emissions), unless they have already been fully
integrated in private costs due to regulatory or legal constraints (liability, carbon tax, safety
standards). Applying this method in the general interest also involves discounting future
factors differently. The authorities’ choice of discount rate is based on notions of equity
discussed above, not on bank interest rates and investors’ demands regarding the rate of
return.
Predictably the disparities between levelized cost estimates are even greater than those
observed between estimates of overnight construction costs, the latter being just one
component of the former. According to the OECD the cost of construction varied by a factor
of one to three between South Korea and Switzerland. In the case of the levelized cost these
two countries still occupy the upper and lower extremities of the range, but with a one-tofive variation in estimates: $29 MWh for South Korea; $136.5 MWh for Switzerland23.
23
A small part of the difference is explained by different discounting values : 5% for South Korea and 10
for Switzerland.
17/36
The values taken into account for the overnight cost of construction and its duration, the
load factor and discount rate explain much of the disparity between the various estimates of
nuclear costs. The cost may be multiplied by four if only extreme, yet realistic, values are
taken into account. Take for example the base case in the 2003 MIT study. The cost per kW
of installed capacity is based on four parameters [$2,000 per kWe; 5 years; 85%; 11.5%].
Taking the extreme values [$2,000 per kWe; 4 years; 95%; 5%], on the one hand, and
[$5,000 per kWe; 6 years; 85%; 12%] on the other, we obtain, respectively, a levelized cost
of $34 per MWh and $161.5 per MWh. The operating costs, including the cost of fuel, weigh
less heavily in the balance, decommissioning and waste-management costs more so. Of
course we are referring here to the cost of next-generation nuclear plants. For aging reactors
nearing the end of their service life, operation accounts for the lion’s share of costs.
Furthermore, decommissioning expenditure being imminent, it adds substantially to costs
unless the operator has already made sufficient provision.
Allowing for external effects does not significantly change the ranking of cost determinants.
According to the simplistic estimate discussed earlier, at the most the risk of an accident only
adds one euro to the average cost per MWh. This is negligible compared with the cost of a
new facility, and low even compared to the cost of operating existing plant. However it is still
only partly internalized, the liability of operators being capped at low levels in the event of
an accident. Nuclear power’s advantage with regard to CO2 emissions could certainly be
taken into account as a social benefit. It could have a substantial impact on the levelized cost
of nuclear power if the price for CO2 emissions was in the upper range (€50 to €100 per
tonne). However it makes more sense to integrate the price of carbon in the levelized cost of
technologies responsible for emissions: indeed, it is integrated through taxes or emissions
permits which directly affect these technologies. We shall consequently examine its impact
when discussing the relative competitiveness of nuclear power.
The curse of rising costs
It is a well known phenomenon that the cost of a technology drops as it is deployed and
becomes more widely used. We have all noticed that we pay less for using a telephone,
computer or airplane than our parents did, simply because the cost of these goods has been
substantially reduced since the first products rolled off factory production lines. Economic
theory cites two causes to explain this phenomenon: the scale effect and the learning effect.
The first one is both familiar and intuitive. The bigger the factory, the less each unit costs to
produce. In other words, the unit cost of large production runs is lower than for smaller
volumes. At the start of a technology cycle the capacity of each production unit is relatively
small, in particular because demand is still limited. Subsequently the size of factories
gradually increases, stabilizing when diseconomies of scale start to appear (due, for
instance, to time spent moving from one workshop to another, or bureaucracy). The learning
effect in manufacturing is linked to the know-how which accumulates over time. The most
intuitive example to illustrate this point is the repetition of a single task. You may spend
more than 10 minutes folding your first paper hen, but barely a minute after making a
thousand or so. Manufacturing an airliner, steam turbine or solar panel is much the same.
The learning effect is generally measured by the learning rate which corresponds to the
reduction in cost when cumulative production doubles. The cost per kWh of wind power
drops by about 10% each time installed capacity doubles24.
24
Asa Lindman and Patrick Söderholm, Wind power learning rates: a conceptual review and metaanalysis, Energy Economics 34, 2012, p754-761.
18/36
Nuclear technology displays the opposite trend. The per-kW construction cost of the most
recent reactors, in constant (inflation-adjusted) euros or dollars, is higher than that of the
first reactors. A technology with rising costs is a very strange beast, which requires closer
study, particularly as this feature distinguishes it from several competing technologies, such
as wind or solar. If nuclear engineering firms fail to find a solution in the near future, the
cost of nuclear power will continue to rise, undermining its competitiveness.
The costs escalation of nuclear power
The rising cost of building nuclear reactors is a well established fact. In particular it has been
studied in depth for installed capacity in the US. The overnight cost of the first reactors, built
in the early 1970s was about $20081,000 per kW. It has increased steadily ever since,
reaching $20085,000 per kW for the most recent reactors, built in the early 1990s. In other
words a one-to-five difference in constant dollars. The increase in the installed cost is even
more striking. The average construction time has increased with time, so interim interest has
increased too. The time taken to build a nuclear power station has risen from between five
and six years for the first plants to be connected to the grid, to more than twice as long for
the most recent units. The average total cost per kWh displays the same upward trend.
Maintenance and operating costs have dropped and the load factor has improved with time,
but these two factors are not enough to counteract the very large increase in the fixed cost
of construction25.
In France the overnight construction cost reported by EDF for its various plants was made
public for the first time in a 2010 report by France’s Court of Auditors26. It amounted to
€2010860 per kW for the first four reactors at Fessenheim and Bugey, commissioned in the
late 1970s, and €20101,440 per kW for the last four reactors, at Chooz and Civaux, which
came online in the early 2000s27. Although it is less than twice the initial amount, the
increase is nevertheless substantial.
Nuclear power consequently has a record of rising costs. But what is the explanation for this
anomaly? A great many factors may have come into play, such as the rising cost of materials
and machinery, or the lack of economies of scale. The figures cited above are the result of
several forces, invisible to the naked eye, which may conceal causes exerting an opposite
force, with varying degrees of influence. To highlight all these factors we need to use a
statistical method known as econometrics. This tool enables us to isolate each of the factors
determining a phenomenon and to measure their respective influence. As early as 197528
econometrics was used to scrutinize the costs of nuclear power in the US. Other work using
the same method has been done since, yielding very interesting results.
Firstly, these works show the absence of any significant economies of scale. The cost per MW
of installed capacity is no lower for the construction of the largest reactors. Why? Because
they are not just scaled up replicas of their predecessors. They are more complex, fitted with
more parts and components, often of a different design. Some research even shows
diseconomies of scale. For instance Robin Cantor and James Hewlett calculated that a 1%
increase in the size of a reactor resulted in a 0.13% hike in the overnight cost per kW. They
demonstrated that initially, other things being equal – in other words maintaining the other
factors they examined at a constant level – the construction cost was significantly less with
lower reactor power (a 1% increase in capacity cuts the cost by 0.65%). However another
25
See the article by Koomey et al on busbar costs.
26
The overnight construction cost for the French fleet, cited in the Cour des Comptes report, amounted
to about €201083 billion. The report also published the construction cost of each pair of reactors, but
these detailed figures do not correspond to the overnight cost, strictly speaking, as they omit
engineering expenses and pre-operating costs.
27
Averages based on Cour des Comptes figures. The difference between Chooz 1 and 2 (€1,635 per
kW), and Civaux (€1,250 per kW) vanishes. The high figure for Chooz is due to the fact that these were
the first units of the N4 series, a new reactor model.
19/36
key factor, construction time, also varies with size. Increasing the size by 1% adds 0.6% to
construction time, entailing in turn a 0.78% increase in cost. The net effect is therefore
0.78 - 0.65, making a 0.13% increase in cost. Large reactors would have been more
economical had they been built as quickly as their smaller counterparts.
Secondly, there were few if any learning effects. This result concerns possible savings for the
nuclear engineering firm. For example, according to Roy Zimmerman29, if the experience
accumulated by a firm rises from four to eight units, it reduces the overnight cost by 4%.
Taking the US nuclear industry as a whole it is difficult to isolate the learning effect
specifically. The figures show that the cost increases with the overall volume of installed
capacity in the US. However this correlation is not due to diseconomies of learning but
rather, as we shall see below, to regulation, which, with passing time, has increased the
construction cost of all reactors. It is important to remember that a correlation does not
necessarily mean there is a relation of cause and effect. There is a correlation between sales
of ice cream and suntan lotion, but one does not drive the other. The correlation is due to a
single hidden variable, the weather, which affects sales of both products.
Thirdly, learning effects appear or are simply greater when utilities act as the prime
contractors on projects, rather than simply purchasing a turnkey plant. There is less
incentive for engineering firms to cut costs. But diminished economies of learning may also
be due to their market power and a better understanding of costs. Firms may take
advantage of their experience to boost profits, to the detriment of their customers. This
conceals learning effects.
Lastly the rising costs are not the result of the accident in 1979 at Three Mile Island, though
it did speed up the process30. The partial reactor meltdown which occurred there delayed
some ongoing construction projects, but the rising costs also concern the overnight cost,
which is not directly impacted by the duration of the project. Furthermore the slowdown in
the US nuclear programme started before the accident. In 1977 the volume of capacity
ordered but subsequently cancelled exceeded built and commissioned capacity. The two
curves crossed over31. The already visible rise in costs partly explains the slowdown in the
US programme.
One variable is missing yet omnipresent: safety regulation. But this variable is hard to
measure, unlike reactor capacity or construction time. The number of texts and their length
is not much use as an indicator, making no distinction between major and minor regulations.
As a result safety regulation is rarely taken into account as a variable in econometric
equations. In 1979 two authors, Paik and Schriver, invented an ad hoc index in an attempt
to integrate regulation. They listed all the regulations issued by the US Nuclear Regulatory
Commission and sorted them into four categories, depending on their supposed importance.
They were thus able to calculate that between 1967 and 1974 regulation had caused a 70%
increase in the investment cost per kW, equivalent to a 16% annual increase. In most other
publications economists have used a temporal milestone (start or end of construction, issue
of building permit) as an approximation for regulation. The work of the NRC continued at a
steady rate all the way through the period during which nuclear plants were being built in the
US; every year it published new standards, rules and measures. The regulation variable may
thus be correlated with time. Any simple variable representing the passing of time, such as
the year when a nuclear plant is connected to the grid, is just as useful as a complex
indicator based on compiling and analysing NRC publications. Using temporal milestones to
inform the regulation variable, US economists estimate that it is responsible for a 10% to
25% increase in construction costs.
28
Bupp IC, Derian J-C, Donsimoni MP, Treitel R, The Economics of Nuclear Power, Technology Review,
p14-25, quoted by Koomey and Hultman.
29
Zimmerman, M. (1982), ‘Learning effects and the commercialization of new technologies: The case of
nuclear power’, The Bell Journal of Economics 13, 297–310.
30
Lucas W Davis, Prospects for Nuclear Power, Journal of Economic Perspectives, vol 26 n°1, 2012.
31
Mark Cooper.
20/36
The inflation in safety regulation is by far the largest factor in the escalating costs observed
in the US. Stricter regulations require larger numbers of safety devices and systems, thicker
containment walls, and completely isolated control rooms. In response to these tougher
requirements engineers design increasingly complex facilities and systems. Only at the end
of the 1990s did it occur to anyone that a possible solution might be to make things simpler,
leading to the Westinghouse’s AP1000, which is based on a passive safety system. Rather
than increasing the number of backup pumps, for instance, a gravity-fed flow would be
maintained if the cooling system failed. In the meantime safety was reflected in higher
construction inputs and overall a more cumbersome framework for coordinating the
construction of plants. The frequent changes in regulations also had a direct impact on the
duration of construction projects. Work on a large number of US power stations had to be
stopped in order to make allowance for new rules introduced since the start of work. Longer
lead times meant higher financial costs, which of course added to the cost of investment.
When new rules required additional inputs, this also impacted indirectly on the overnight
cost. And, despite it being based on the assumption that plant was built in one night, longer
lead times pushed up overnight costs in the US.
At first sight analysis of the escalating costs of nuclear power in the US might suggest that
stricter safety requirements imposed by the regulator are to blame. But several factors
contradict such a simplistic conclusion. It is not so much the severity of regulation as its own
defects that cost US nuclear power so dearly. Fluctuating rules and shifting priorities,
excessive delays in decision-making and an inadequate understanding of the fundamental
technical issues may generate excess costs for utilities, which far outstrip the impact of rising
safety requirements. It seems more probable that, up to the end of the 1970s, the
regulations did not so much attempt to raise the initial safety level as simply to achieve it. It
is far from easy to assess the safety level of a nuclear power station, particularly before the
fact, simply on the basis of drawings. Building and operating a plant may ultimately reveal
that it does not meet the safety targets set by the regulator, and the operator, at the design
stage. So the regulator intervenes to ensure that the original safety targets are fulfilled. This
may remedy defective quality but does not raise its level. Some authors, such as Mark
Cooper, assert that early US reactors were quite simply defective in safety terms and that
regulation imposed a form of making good. Lastly, if we read between the lines of escalating
US costs we may detect serious shortcomings in terms of industrial organization. Divided
into a large number of utilities, often small and limited in territorial reach, and a host of
engineering firms, the industrial organization failed to achieve sufficient standardization of
procedures, reactor models and construction practices. Apart from Bechtel, which built 24
reactors, the experience of engineering firms and operators was limited to building just a few
nuclear plants. In short, unlike many other fields of technology in which the US led the way,
the development of nuclear power on an industrial scale was not a great success.
The picture in France was very different, whatever its critics may have maintained (see box).
It has now been firmly established32 that the escalation in costs was far less spectacular, with
overnight costs rising by 1.7% a year, compared with 9.2% in the US.
A dizzy rise in costs based on mistaken analysis
In 2010 an academic journal published an article33, which attracted considerable attention.
For the first time the construction costs of French reactors were detailed and tracked over
time. But contrary to what everyone imagined, the figures showed that France, despite its
assets, had also suffered a steep escalation in costs: the cost of building France’s last four
reactors was allegedly 4.4 times higher than that of the first four. Worse still, the last reactor
to completed (Civaux 2) purportedly cost 7.5 times more than its cheapest counterpart
32
See Lina Rangel Escobar and François Lévêque, 2013.
33
The Costs of the French Nuclear Scale-up: a Case of Negative Learning by Doing, Arnulf Grubler,
Energy Policy 38, 2010.
21/36
(Bugey 4). It seemed that through some intrinsic fault nuclear technology was incapable of
controlling costs and impervious to learning effects. The large scale of the construction
projects, the limited unit count, the need to adapt to different sites, and the task of
managing such a complex undertaking all contributed to cancel out the cost-cutting
mechanisms observed elsewhere: standardization, production runs comprising several
thousand units, and the repetition of almost identical processes.
This diagnostic would have been justified, had it not been founded on a mistaken estimate.
In the absence of publicly available data on the construction costs of each French reactor,
the author of the article, Arnulf Grubler, extrapolated the cost of plants from EDF’s annual
report on investments. Work had been carried out on several reactors – often of different
sizes – in the course of the same year, so Grubler had broken down annual investment,
using a theoretical model of expenditure to estimate the cost of each plant. Unfortunately
this extrapolation yielded figures which subsequently proved to be at odds with reality. Far
from a more than fourfold increase in the construction cost of reactors, from start to finish,
the data later published by the Court of Auditors revealed a slightly less than twofold
increase, in no way comparable to what had happened in the US.
So why was there such a big difference between the United States and France?
Econometrics is unfortunately not much help here. On the one hand, only a small amount of
work has focused on France’s nuclear reactors; on the other, the sample itself is small. In all
we only have 29 records of costs. France has a total of 58 reactors, but they were built in
pairs and the EDF accounting system did not itemize them separately. With such a small
sample, fewer variables can be tested. With respect to economies of scale, there is no sign of
a positive effect, quite the opposite. The nameplate capacity of French reactors increased in
three steps, rising from 900 MW for the first reactors, through 1,300 MW for the majority of
them, culminating at 1,450 MW for the last four. It is immediately apparent that the cost per
kW went up with each step or palier, with a particularly spectacular leap at the end. The
overnight cost reached €20101,442 per kW, compared with an average of €20101,242 per kW
for the 20 second-step reactors, or €20101,121 per kW for the first 54 overall. Econometric
analysis yields no further information on this point; the diseconomies of scale persist. Here
again the explanation is to be found in the relation between size and complexity. Not only
did the reactors on each step differ in size, they varied in other ways. Each step brought
technological advances. For example the second-step plants were equipped with a
completely updated control room and system. The design of the last four was almost
completely different. When it comes to learning effects, econometric analysis is more helpful,
revealing that the overnight cost of a reactor fell depending on the number of reactors
already built on a given palier. Each additional reactor brought a 0.5% drop in cost. On the
other hand the effect is no longer visible if we look at the total number of reactors previously
built. Apparently the experience gained building one model of reactor did not benefit a
different model.
It is essential to grasp the step-related learning effect, because it throws light on a recent
controversy. The French nuclear programme offered the best possible conditions for powerful
learning effects. The power stations were built by a single operator, EDF, which was able to
appropriate all the experience accumulated with each new project. The plants were built in a
steady stream over a short period of time. In the space of just 13 years, from late 1971 to
the end of 1984, work started on construction of the first 55 reactors. The programme as a
whole only slowed down at the end, with work on the last three units starting between late
1985 and mid-1991. The average construction time was consistent, only increasing slightly
over time. Unlike what happened in the US, the regulatory framework did not upset
construction of nuclear plants. The fleet expanded gradually thanks to dialogue and
22/36
cooperation between all the players (EDF, Atomic Energy Commission [CEA], Framatome,
Ministry of Industry), well out of sight of non-specialist outsiders.
So, despite the fact that France enjoyed the most favourable conditions for a gradual drop in
the cost of building nuclear power plants, this did not materialize. What went wrong? We
may suggest a series of specific explanations: the easiest sites were chosen first; quality
assurance was gradually tightened up; the rising price of energy impacted on the price of
machinery; project ownership expenses increased34. At a more fundamental level, the French
nuclear programme was over-ambitious and focused too much on just one country. The
standardization and learning effects it made possible were cancelled out by changes in
reactor models. The two capacity increases, from 900 MW to 1,300 MW, and then from
1,300 MW to 1,450 MW, coincided with substantial, expensive changes in technology. Some
were adopted to make the technology French. In an effort to achieve greater independence
and improve its chances of exporting its own reactors, France was determined to break free
from the US technology used in the first pressurized-water reactors built there. The first
stage in this process involved the design of the P’4 variant of the first-step 900 MW reactor.
This dispensed with the need to pay licence fees to Westinghouse. The second stage brought
the original design of a 1,450 MW reactor, but ultimately only four units were built. This
model proved more expensive than its predecessor, due to its greater technological
complexity and the exclusive use of components and machinery made in France35. In
addition construction times grew longer, reaching an average of 126 months for the last four
plants, half as much again as for the plants built during the previous step. The French
nuclear programme was nearing its end, indeed rather sooner than expected, because
growth in demand for electricity, with a corresponding increase in capacity, had been
overestimated. Completion of the last reactors was deliberately spread out in time, to adjust
to demand and cope with the gradual winding down of the workforce [caused by the end of
the construction programme]. Things are always clearer with the benefit of hindsight, but it
does look as though France could have done without the last four reactors, yielding a
substantial saving.
Together the US and France have a total of 162 reactors, equivalent to just under a third of
global capacity. What is known about the costs of other reactors? Nothing! There is no public
source of data for all the nuclear capacity deployed in the former Soviet Union, Japan, India,
South Korea or the People’s Republic of China. No figures are available to say whether costs
escalated there too, less still at what rate. We can only resort to qualitative reasoning. Apart
from South Korea, no doubt, and perhaps China more recently, it is hard to imagine costs
rising less than in France. South Korea enjoys similar conditions, which should have enabled
costs to be contained: swift pace of construction; reasonably similar reactor design and
layout; well integrated industry and a single operator; nationalist fervour. In fact it may have
done better than France. The picture in China is much more disparate, featuring all types of
technology – boiling water, pressurized water, heavy water – and many sources – Canada,
Russia, France and even the US. Less than 10 years ago China decided to give priority to
building large numbers of its own CPR-1000 reactor, derived from the French 900 MW model.
The speed of construction has been stupendous, great efforts have been made to standardize
processes and the industry is very well organized. The cost of building this reactor has
probably dropped with each new unit.
On the other hand the former Soviet Union and India would be plausible candidates for
notching up escalating costs even worse than in the US. In the first case because costs under
the socialist system were never a key issue when deciding to invest in infrastructure. Politics
had more say than economics in the siting of plants, in the choice of model and the speed of
construction. India is well placed too, no country having witnessed such a chaotic civil
nuclear programme.
34
Bertel and Naudet, 2004 and (quoted in the work of) Moynet,1984.
35
Grubler 2010.
23/36
Is there no limit to escalating costs?
Will what happened yesterday hold true tomorrow? We are confronted with a classic case of
inductive reasoning. We have seen that the second reactor costs more than the first one, the
third one more than the second ... and that reactor n costs more than n-1. So can we
conclude that the same progression will hold true for n+1 and n+2. The immediate answer is
affirmative. If you have only seen black cats in the past, you will be quite ready to bet they
are all black. In the past nuclear power has reported rising costs, so nuclear technology is
synonymous with rising costs. It is tempting to generalize. Particularly as new nextgeneration reactors – the ones following the nth reactor such as the EPR – are again more
expensive than their predecessors. However, we shall see that it is possible to upset this
progression, even if it is much less likely than the continuation of the previous trend.
Research would also need to explore new routes, with industry finding ways of standardizing
models and developing modular machinery. If no spell is found to lift the curse of escalating
costs, nuclear power will be gradually sidelined.
At the beginning of the 2000s costs seemed to have stopped escalating. Next-generation
reactors were expected to bring improved safety, but they would also be cheaper than their
forebears (see box). On paper the outlook for nuclear costs was rosy, on both sides of the
Atlantic.
Costs at renaissance
After a long, sluggish period in western countries, nuclear power woke up again in the early
2000s. New construction projects were tabled in the US and Europe. Many countries with no
previous experience of nuclear power were also eager to enter the technological fray. This, it
seemed, marked the so-called renaissance of nuclear power. The International Energy
Agency forecast the construction of several hundred new plants by 2030. The outlook on
costs was naturally just as upbeat. In 2003 the MIT published a study estimating the cost of
building a plant with a next-generation reactor. In its base case it assumed an overnight cost
of about $2,000 per kW, which yielded a levelized cost of $67 per MWh (with an 11.5%
discount rate). To situate the latter cost in relation to the past36, let us imagine a scale of 1
to 100 ranking existing US plants by rising cost (calculated in constant dollars, adjusted for
inflation and with a uniform 6% interest rate37). The MIT’s projected plant would be ranked
19th, in the top 25% least expensive plants ever built, reaching back to the 1970s. In an
even rosier scenario, positing a swifter, more flexible response by administrative bodies for
the issue of construction permits, the cost would be lower than any plant previously built in
the US. A year later the University of Chicago carried out a similar study, drawing
comparable conclusions. On the supply side Westinghouse announced an overnight cost for
its AP1000 of $1,400 per kW38 and a levelized cost of $27 per kWh39. Predictably this
estimate was more optimistic than the ones produced by university research laboratories.
In France the baseline costs were published by the Ministry of Energy. In 2003 the costs for
third-generation nuclear plants were estimated at €1,300 per kW for the overnight cost40 and
€28.4 per MWh for the levelized cost (with an 8% discount rate). With these values the EPR
bettered, in terms of cost, the reactors on the last step built in France. Industry was slightly
less optimistic, with EDF suggesting an overnight cost of between €1,540 and €1,740 per kW
and a levelized cost of €33 per MWh41.
36
Koomey and Hultman, 2007.
37
With a 6% discount rate, the levelized cost in the MIT study is $42 per MWh.
38
Zaleski, p3, first of a kind, Chine, p. 3,
http://nuclearinfo.net/Nuclearpower/WebHomeCostOfNuclearPower
39
Koomey.
40
Glachant and Lévêque (ed), Electricity Reform in Europe. Towards a Single Energy Market, Edward
Edgar, London, 2009.
41
Dupras, Joudon, Revue Générale Nucléaire, VI-2004, and Zaleski.
24/36
Barely 10 years later, the first construction projects soon showed that the de-escalation
everyone hoped to see had not yet started. The next-generation reactors were even more
expensive. Present trends are after all entirely consistent with those of the past.
In 2009 the MIT published a second report42, updating the findings of the initial study six
years earlier. The increase in the overnight cost was spectacular: expressed in current
dollars it doubled, rising from $2,000 to $4,000 per kW43. In particular this figure took into
account the estimated costs of 11 projected plants in the US, for which the relevant utilities
had applied to the regulatory bodies for reactor licensing. Meanwhile the University of
Chicago investigated applications for construction licences for the Westinghouse AP1000. On
average, the overnight cost quoted in applications was $20104,210 per kW, multiplied by a
factor of 2.3, in constant dollars, compared with a study seven years earlier44.
Unlike what occurred in the US, where next-generation reactors went no further than the
drawing board, construction projects in Europe got off the ground. Work started on two
EPRs, one at Olkiluoto, Finland, the other at Flamanville, France. Here the increase in costs
has been even more spectacular. In Finland the initial cost of the project when work started
was €3 billion45, or €1,850 per kW. It has since been revised upwards on several occasions;
delays have accumulated too. The final cost is now estimated at €6.6 billion, or €4,125 per
kW. The job was supposed to last four and a half years, with grid connection in mid-2009. In
the end, production will not start before 2014, at best. Say 10 years to be on the safe side.
Work at Flamanville started two years later and took the same unhappy route as its elder
sister. The initial cost of €3.3 billion46 has soared to €8.5 billion47 and the original
construction time of under five years will probably stretch to nine years. So the first EPRs
cost much more than the preceding 1,450 MW reactor model, on which they are based.
The changes in academic studies and industrial quotes are so large that it would be easy to
make fun of them, or even to suspect deception. But it would be a mistake. It is only natural
that the initial estimates of experts and vendors should be a little optimistic. But for new
nuclear there were neither experience nor facts to temper initial optimism. After a long
period without any new plant being built, a large share of American and French expertise had
vanished. Most of the engineers and senior executives who had taken part in the golden age
of nuclear power had either moved to another sector or retired. Furthermore the first cost
estimates were drafted when design of the next-generation reactors was still in its early
stages. Millions of man hours were still needed to finalize detailed plans48, which inevitably
revealed additional costs. Then it was time to obtain quotes from suppliers and to sign
contracts for parts and machinery, a process which moved the true understanding of costs
one step further. The last set of estimates generally focuses on indexed values, in particular
the price of raw materials and building materials. This brought additional price increases, the
first decade of the 2000s having seen substantial upward pressure on these commodities.
The overnight cost of gas and coal-fired power stations also increased steeply over this
period49. The difference with nuclear power was that the initial estimates for the fossil-fuel
42
Update of the MIT 2003 Future of Nuclear Power, MIT, 2009.
43
Expressed in $2007, with a 25% increase in the levelized cost.
44
The lower end of the overnight cost for the 2004 study was $20101,413-2,120 with a mean value of
1,765, hence the increase by a factor of 2.3. Focussing just on the AP 1000, the 2004 range was
$20101,554-2,331 with a mid-point at $20101,943. So the study took $2,000 per kW and compared this to
$4,210 per kW, in other words an increase by a factor of 2.1.
45
According to French Member of Parliament Marc Goua, tasked with reviewing the accounts of Areva
and EDF, http://www.enerzine.com/2/12796+lepr-finlandais-couterait-au-final-6-6-mds-deuros+.html,
14 October 2011.
46
Le Monde, http://www.lemonde.fr/planete/article/2011/11/10/sur-le-chantier-de-l-epr-a-flamanvilleedf-est-a-la-moitie-du-chemin_1602181_3244.html.
47
EDF communiqué cited in the Cour des Comptes report.
48
See second study by the University of Chicago.
49
For example, in its updated study the MIT revalues the overnight cost per kW of a gas plant, resulting
in a 70% increase, and a 130% increase for a coal-fired thermal plant.
25/36
plants were more accurate. They were based on a building process which had never stopped,
nor yet slowed down, all over the world, with hundreds of examples on which to draw.
Optimism may also be dictated by self-interest. Utilities in favour of nuclear power and
reactor engineering firms stand to gain by reporting low costs in their initial estimates, by
only publishing values at the lower end of their spread estimates. But on the other hand,
much as any trader selling goods to a small number of buyers, on whose custom the
business depends, it is not in the interest of reactor vendors and turnkey plant integrators to
announce miraculous figures. Making promises, which they know they cannot keep,
permanently saps their credibility in the eyes of customers, bankers and governments. If
there was any deceit regarding costs at the renaissance of nuclear power, it was the industry
which fooled itself.
To put an end to any notion of cross-the-board deceit, it should also be borne in mind that
the baseline academic studies did not only work on a set of assumptions favourable to
nuclear power. The reason why the first MIT study caused such a stir in 2003 was that it
made the iconoclastic choice of a high discount rate, which was unfavourable to nuclear
power. The MIT highlighted the high financial risk associated with this investment in
liberalized electricity markets. As a result, the assumptions regarding the structure and cost
of nuclear capital were less attractive than for gas or coal. Nuclear power involved higher
capital outlay, less debt and a 15% return on assets, rather than 12%. Without these
assumptions the MIT study would have concluded that the excess cost of nuclear power,
compared to gas, was only half as large50.
There is no escaping the facts and they are particularly stubborn: nuclear power now is much
more expensive than before. For the time being third-generation reactors are still plagued by
rising costs, and new reactor models bring additional costs. What does the future hold?
With the same design, costs should certainly drop, but by how much? It is impossible to say
whether there will be a slight reduction or a huge one. Take the EPR. Its cost is bound to
drop, but how far? First of a kind costs are known to be higher, generally by about 20% to
30%51, but it is not known how the excess cost is amortized. Does the full burden fall on the
first unit, or is it spread over the first five or 10 reactors? For obvious reasons – the first
customers do not like teething problems – data of this sort is confidential. Furthermore there
has been a loss of experience on the construction side, following a long period without any
new projects. Lastly, the first two EPRs are not being built by the same company. Seen from
abroad, the French nuclear industry may look like a homogenous block; EDF and Areva, both
publicly owned companies, seem barely distinguishable. But in fact they have been keen
rivals in recent years. Areva went it alone in Finland, operating as a turnkey plant vendor,
rather than just selling a reactor, which is its core business. EDF has longstanding experience
as both the prime contractor and project owner of nuclear plants. It sees Areva as an original
equipment manufacturer, or even – rather disparagingly – as a boiler manufacturer. So there
is no sign of any learning effects between Olkiluoto and Flamanville. The two firms have been
at loggerheads, rather than pooling their experience. The opposite seems to have happened
at Taishan, in China, were two EPR-powered plants are being built. EDF and Areva are
working together with the prime owner, the China Guangdong Nuclear Power Group, the
utility in Guangdong province. For the time being Taishan-1 is on target for both construction
time (five years) and cost (€3 billion). Areva management52 say this is thanks to the return
on experience from the Finnish and French jobs. Certainly, between Olkiluoto and Taishan,
the supply deadlines have improved by 65%, engineering man hours for the nuclear steam-
50
The levelized cost for the base case was $67 per MWh for nuclear, $43 per MWh for coal and $41 per
MWh for gas. With the same financial conditions the cost of nuclear power drops to $51 per MWh. See
Table 1, Yangbo Due and John E Parsons, Update on the Cost of Nuclear Power, MIT, May 2009.
51
The 2004 University of Chicago study suggests that First Of A Kind costs may be as much as 35% of
the overnight cost. For their part Dupraz and Joudon, cited by Zaleski, estimate that FOAK costs add
20% to the levelized cost of the first of a kind, for a series of 10 units (€200441 per MWh, instead of
€200433 per MWh).
52
Luc Oursel
26/36
supply system are down by 60%, and the time taken to build the main components has been
cut by 25% to 40%. So the third reactor seems poised to finish first. Work on Taishan-1
started in 2009, after the other two, but it should be connected to the grid by the end of
2013, several years ahead of Olkiluoto and Flamanville. But return on experience is not the
only reason for the impressive performance in China regarding costs and deadlines. The
Public Republic of China boasts top-notch civil engineering contractors, can count on a
seasoned nuclear industry, is deploying a massive programme (with 26 reactors under
construction in 2011), and has the advantage of a cheap, well qualified workforce and a well
organized site where work continues round the clock, even at weekends.
The last unknown regarding the scale of the drop in the cost of the EPR relates to the
number of units ultimately built worldwide. Four, 10, 20 or more? All other things being
equal, the more reactors sold, the lower the cost and vice versa. The serpent eats its tail.
Potential buyers are price-sensitive – though we do not know whether this effect is very
slight or substantial – and learning effects cut costs, though here again we cannot say by
how much.
From a technical point of view the key to lower costs is to be found in standardization and
modularity. Standardization requires every unit of a particular reactor model to be identical,
which is not always the case, due to specific changes demanded by customers or safety
authorities. As mentioned above, standardization allows learning effects; we may add that it
also facilitates competition between suppliers, another powerful mechanism pushing costs
down. Modularity means construction in modules, in other words component parts which are
relatively independent one from another, making it easy to separate them and simply
assemble them on-site (structural elements, but also cable ducts, reinforced concrete mats,
etc.53). A good example of modular building is factory-assembly of the roof timbers of a
detached house, rather than erecting them piece by piece on-site. Pre-assembly is
advantageous because a factory is a sheltered environment and such operations lend
themselves to automation, yielding productivity gains. Pre-assembly also reduces the
amount of clutter on a building site, streamlining its organization. So modularity has the
potential for substantial gains54.
So far our reasoning has been based on an unchanging technological framework. What
happens to the costs entailed by nuclear power if we take into account innovation, and the
design and development of new reactors? Past form is far from encouraging. We have seen
that in France, where conditions were most favourable, each new model led to an increase in
the construction cost per kW of installed capacity. Two insurmountable obstacles seem to be
preventing a reduction in the cost of new models. The first relates to the increasingly strict
rules on safety. It is hard to imagine the authorities certifying a new model with lower safety
performance than its predecessors. As time passes experience gained from building and
operating plants reveals defects; progress in science and technology provides solutions to
correct them. Furthermore, with time, new political risks may emerge (terrorist hijacking of
an aircraft to target a power station, for instance) and in general public opinion is
increasingly averse to technological risks. The above is true for countries already equipped
with nuclear power. For new players safety requirements may be less stringent and they
may not require the latest generation of reactors. But keen to develop their science and
technology, such countries are unlike to resist the appeal of modernity for long.
So the question is whether it is possible to build reactors which are similar to the current
generation, but safer and cheaper. Very probably not, but as it is still too soon to pass
judgement on the AP1000, we should allow for a positive outcome. Westinghouse designed
this reactor with two aims: to provide a mechanical solution to some of the safety problems;
and to simplify the overall design. For example, water tanks are positioned on the roof in
order to cool the reactor vessel should the need arise, fed by gravity and the pressure inside
53
Reduction of Capital Costs of Nuclear Power Plants, OECD, February 2000.
54
Occasional, initial and old experience of construction in Sweden and Canada estimated at between
1.4% and 4%. See OECD study, 2000, p10.
27/36
the system. This more or less halves the need for pumps, valves and pipework. Four
AP1000s are currently under constructionin China. It will be interesting to see, in a few
years’ time, whether they cost substantially less to build than the EPR. If the concept is a
success, it could lead to the development of improved versions, using the new design rules,
but at even lower cost. Nuclear power may finally cast off the curse of rising costs.
The second, apparently insurmountable obstacle concerns on-site construction and short
production runs. Much as other large civil engineering projects – bridges, airports or dams –
nuclear power stations are mainly built on-site. Progress may be made towards greater
modularity, but there is little hope of a 1,000-MW plant one day being put together like a
flat-pack kitchen. Civil nuclear power differs from other electricity-generation technologies in
that only a small number of units are built. Whereas hundreds or thousands of wind-farms,
or coal or gas-fired plants are ordered worldwide every year, there are just a few dozen new
construction nuclear construction projects. One of the reasons is the trend towards building
increasingly large reactors. The scale of fixed costs justifies this option, because they can be
recouped on a larger volume of electricity output. But there is nevertheless a downside. All
other things being equal, the more powerful the reactor, the smaller the number of identical
units built. So production runs are short and only a few similar parts and components are
manufactured. The trade-off between economies of scale per unit and manufacturing
economies of scale55 has so far tipped in favour of the former. Giving fresh impetus to smallreactor projects would break with this approach.
The example of small reactors is worth looking at, because it demonstrates the scope for
radical innovation, which in our opinion offers the only lasting antidote to the curse of rising
costs. People have been developing low-power nuclear reactors for many years. They are
used to drive nuclear submarines, drawing on work and trials going back to the 1950s. What
is new though is the sudden emergence of futurist projects. Take for instance the best
known example, funded by Microsoft-founder Bill Gates. The project is being developed by
TerraPower, in which he is the prime shareholder. The aim is to produce a mini-reactor
several metres high, running on natural uranium and cooled by liquid sodium. It is based on
the travelling-wave principle, with the reaction slowly spreading outwards from the core of a
block of uranium. Picture a candle with a flame inside gradually advancing as it consumes
the surrounding wax. For the reactor itself, imagine a cylinder less than one metre high,
which requires no outside intervention once the reaction has started and which shuts down
on its own after several tens of years. We may also cite the project for an underwater
nuclear power station being developed by France’s naval defence firm DCNS. In this case the
cylinder is 100 metres long and 15 metres in diameter, containing a reactor and remotecontrolled electricity generating plant. With several tens of MWs capacity, it would be located
out to sea, several kilometres from the coastline, anchored to the seabed. The cylinders
would be modular units, several of which could be placed side by side, in the case of higher
output requirements. The units would be taken back to a shipyard for maintenance and
replaced by other units, much as bottles with a refundable deposit. These projects, which
sound even more fantastic when described in such brief terms, will very probably never see
the light of day. Either they will founder completely or change so much that the final
application bears no resemblance to the initial concept. It matters little to our current
concerns. That is how radical innovation works: projects pursuing a large number of original
ideas are launched; very few give rise to pilot schemes; an even smaller number lead to
commercial projects; and in each case the ongoing redefinition process will shift pilot
schemes and commercial goods further and further away from the original idea. Obviously
there is no way of knowing in advance whether, out of the hundreds of current and future
projects to develop modular small or mini-reactors similar to those discussed above, at least
one could reach fruition and enter industrial production. But unless nuclear research moves
away from the present model of large, non-modular plants and gigantic construction
55
Nemet, 2007, quoted by Koomey.
28/36
projects, the costs of nuclear technology will continue to rise, which is a serious drawback in
the competition between nuclear power and other electricity-generating technologies.
Nuclear power and its alternatives
We cannot do without oil but we may, on the other hand, stop using the atom. We should
never lose sight of the fact that there are several means of generating electricity, using
among others coal, gas, oil, biomass, solar radiation and wind. At the scale of a whole
country these generating technologies are generally combined to form an energy mix, which
may or may not include nuclear power, much as it may or may not include thermal coal or
gas, wind or solar.
The various technologies are both competitors and complementary. Conventionally a
distinction is made between base load generating technologies, coal or gas-fired powered
stations for example, which operate round the clock all year long, and peaking generating
technologies, such as oil-fired power stations, which only operate at times of peak demand.
With a finer mesh, a distinction may sometimes be made between semi-base load and
extreme-peak generation. The overall idea is to classify production resources in such a way
that the ones with high fixed costs and low variable costs are used for as many hours a year
as possible, while on the other hand those with low fixed costs and high variable costs are
only used for a few hours a year. We shall analyse this rationale in detail in another paper.
However it is immediately apparent that two categories of base load technology – coal and
nuclear – are in competition, whereas oil-fired technology is complementary. However, in
situations where they overlap this ranking may change. For example gas, which tends to be
seen as a semi-base load resource, may play a primary role as a base load resource; nuclear
power may lend itself to load-balancing and is consequently suitable as a semi-base load
resource. Renewable energy sources also upset the ranking. Hydro-electric power from dams
is generally seen as a peaking resource, despite its extremely high fixed cost and variable
operating cost close to zero, the explanation being that its variable cost should in fact be
treated as a marginal opportunity cost. It is preferable to hold back a cubic metre of water
for peak hours with correspondingly high prices, rather than wasting it by generating
electricity at times when demand drops and the price is low. Regarding wind and solar,
production is intermittent when it depends on the force of the wind or the amount of
sunlight, which vary in the course of a day, and from one day to the next, quite beyond our
control. Here again variable technical costs are close to zero, but the irregular nature of
output makes it impossible to classify these technologies among base load resources. At the
same time, the lack of any way of controlling them means they cannot be treated as peaking
resources. If intermittent renewable energy sources play a significant part in the energy mix,
backup capacity must be available – generally gas-fired plants – to take over in the absence
of sunlight and wind. Under these circumstances gas and the renewable energy are
complementary. On the other hand the growth of intermittent energy sources pushes the
market price of electricity down and base load and semi-base load sources operate for
shorter periods. This creates competition between nuclear power and gas, on the one hand,
and renewable energy sources, on the other. Lastly nuclear power and renewables have one
characteristic in common: they produce no CO2 emissions. They may consequently be seen
as rivals for achieving the targets set for reducing greenhouse gas emissions, or
alternatively, as it seems difficult to rely exclusively on just one of these sources, they may
be seen as complementary, with a view to completely carbon-free electricity generation. To
simplify matters, any comparison of nuclear electricity should make allowance for two
factors: on the one hand its competitive or complementary position in relation to coal or gas,
for base load electricity production; and on the other hand its competitive or complementary
position in relation to other carbon-free energy sources.
29/36
The relative competitive advantage of nuclear power over gas or coal
The levelized cost enables us to classify the various generating technologies. Which one, out
of coal, gas or nuclear power, offers the lowest cost? How do these forms of energy rate in
the overall cost ranking? Our obsession with rank prompts us to ask the wrong questions,
which only yield contingent answers.
There is no single ranking system because the costs depend on different locations and
hypotheses on future outcomes. With regard to nuclear power we have seen that the cost
varies from one site to another, from one country to the next, and that it above all depends
on the discount rate. The cost of fuel is the key parameter for coal and gas. But the price of
energy resources depends on geography. The cost of transporting coal or gas being high,
building a fossil-fuel power station in one place or another yields different results.
Furthermore market prices fluctuate a great deal, particularly for gas, often indexed on the
price of oil. The rate of return on an investment in a new fossil-fuel plant depends on
assumptions as to how fuel prices will behave over the next 10 or 20 years. Consequently it
is only possible to use the levelized cost to rank coal, gas or nuclear power on the basis of a
very specific set of conditions, valid at the geographical scale of a country and in line with
the expectations of specific operators. For example, taking a broad-brush approach to the
current position in the US, gas enjoys a comfortable lead, followed by coal, with nuclear
power in third place. This ranking may vary between US states depending on the proximity
of coal-mining resources and unconventional gas reserves.
We should nevertheless bear in mind a few, almost universal trends and shifts, which also
happen to explain to a large extent the current US ranking of base load generating
technologies: before and after climate-change policy; before and after shale gas; before and
after deregulation of the electricity market.
In a world with no pollution-abatement measures, coal would lead the pack with the
cheapest MWh almost all over the world. But using it to generate electricity causes local
pollution (release of dust, soot, sulphur and nitrogen oxides) and CO2 emissions. The first
group is by far the most costly, unless a very high price is set for CO2 (in excess of $100 per
tonne)56. In ExternE, the major European study of the externalities of generating electricity,
the damage caused by coal, setting aside that linked to CO2 emissions, was estimated to
range between $201027 and $2010202 per MWh. The lower value in this range is the same as
the one reported by William Nordhaus and other authors in a conservative assessment dating
from 201157. As for the upper value, it can be found in a maximalist study by Professor Paul
Epstein, at Harvard, published the same year58. Taking the values which the experts consider
to be the ‘best estimates’, we may note that the cost of a coal-generated MWh doubles when
we include its externalities. The large divergence between the upper and lower values in the
estimates can be partly explained by the different types of plant under consideration and the
prevailing environmental standards. In OECD countries the regulatory framework for local
emissions from coal is very strict. Part of the externalities is internalized by emissions
standards, which raises the overnight cost of coal-fired thermal plants, and consequently the
levelized cost of energy for the utility. Similarly some OECD countries have introduced a
carbon price, or are planning to do so. Depending on their level, such taxes and tradable
emissions permits internalize, to a greater or lesser extent, a share of CO2 externalities and
add to the variable cost borne by the utility. On the other hand, in most developing or
emerging countries, the cost of a coal-generated MWh is still low because neither investors
nor utilities pay for any part of the environmental damage entailed, in the absence of both
regulations on local pollution and a carbon price. This lack of symmetry explains why it is
now almost out of the question to build coal-fired power stations in the US, the UK or Japan,
56
[note manquante ?]
57
Value reported as $28.3 per MWh. Article co-authored with Nicholas Muller and Robert Mendelsohn,
American Economic Review.
58
Full cost accounting for the life cycle of coal. Here the value was $269 per MWh, of which $44 per
MWh corresponds to the impact on public health.
30/36
whereas such facilities are springing up in China, Malaysia, Senegal and South Africa. In
terms of new electricity-generating capacity being installed, coal is by far the technology
which has enjoyed the strongest growth worldwide since 2000. In the long term, the cost of
a coal-generated MWh in non-OECD countries is expected to rise, reducing the gap. The
localized pollution and damage this technology entails for public health exert pressure which
encourages a shift towards other more expensive technologies which cause less pollution. In
OECD countries it is more difficult to predict future developments. The application of R&D
work on clean coal, particularly for carbon capture and storage technology, is uncertain.
Future trends for the price of CO2 emissions are equally uncertain.
Gas has a very different environmental profile from coal, with little or no local pollution, and
half the volume of CO2 emissions. This explains its spread in OECD countries, at the expense
of coal. The price of gas delivered to the generating plant is generally higher than for coal,
but this competitive disadvantage is counterbalanced by incomparably lower environmental
costs59. There is certainly a before and after unconventional gas here, because this
advantage is now being enhanced by lower costs due to new gas-exploitation techniques
(horizontal drilling and hydraulic fracturing), and the resulting extension of reserves. In the
US, where shale gas was first exploited (alongside Canada), this change means that nuclear
power is durably losing its status as a base load generating technology. Gas is now in first
place and is likely to stay there for a long while. However it should be borne in mind that
unconventional gas currently enjoys a novelty effect, which underestimates its social cost. It
took decades to work out the economic estimates of the externalities of coal, conventional
natural gas and nuclear power. They took shape as science advanced in its understanding of
the effects of pollution and on-site measurements. The dissemination of scientific advances
and the results of metrology, beyond the confines of laboratories and a small number of
experts, works on a specific time scale. None of this applies to shale gas, yet. The
measurements and studies have barely started, particularly to estimate greenhouse gas
emissions and possible damage to aquifers. It is plausible to suppose that what has so far
been gained through lower exploitation costs may tomorrow be lost to rising environmental
costs. Lastly it is worth noting that the decision by some markets to delink oil and gas prices
gives the latter an advantage which is likely to last. Until now, in many countries gas prices
were driven up by the rising price of oil. Oil-indexed gas supply contracts were encouraged
by various factors: comparable extraction conditions; joint production in some cases; and
markets offering imperfect competition, due to the dominant position of monopsonists. In
places where the exploitation of conventional gases has developed, this arrangement has
been permanently destroyed.
Liberalization of the gas and electricity markets is the third key shift which changes the
relative competitiveness of base load generating technologies. Here too nuclear power has
lost ground on the whole. For many years the gas and electricity markets were organized as
municipal, regional or national monopolies subject to regulated tariff schemes. Regardless of
whether generating companies belonged to the public or private sector, the investments they
made were exposed to little risk, being paid back by captive consumers. Dependent on the
authorities, these companies often acted as cogs in the implementation of energy policies
based on factors related to cost, but also to national independence, scientific prestige, job
creation and such. Instigated by some US states and the UK, privatization and the opening
up of the gas and electricity markets to competition upset this model. In its place, or
alongside it, another model was established in which the link between production and captive
consumption was broken, and in which investment was decided by shareholders and
bankers. From being utilities – public service providers – the electricity generating companies
became operators at the head of merchant plants, power stations selling electricity to the
wholesale market. The risks here were not of the same order. Much as football teams which
59
The ExternE study estimates the external costs of electricity generated using natural gas (excluding
carbon emissions) at between $201013.4 and $201053.8 per MWh, as against $201027 to $2010202 for coal
(reported in The Social Cost of Coal, Samuel Grausz, October 2011, Climate Advisers). Natural gas
produces half as much carbon emissions as coal.
31/36
compete on the same playing field, be it muddy or too hard, one might suppose that
liberalization would affect all the electricity generating technologies in the same way.
Accordingly the new deal should not alter their competitive positions in relation to one
another. In practice this did not prove to be the case for nuclear power, which, as far as the
financiers were concerned involved greater, more serious risks60: higher risks of budget
overruns and missed deadlines, in the course of construction and during operation (e.g.
safety defects leading to unpredictable reactor shutdowns and consequently lost output); a
long period over which to recover investment, increasing the risk due to uncertainty in
wholesale electricity markets; higher regulatory and political risks due to the opposition of
part of public opinion and some political parties to atomic energy. In the face of these
additional risks, the MIT study cited above set a weighted average capital cost 25% higher
than for gas and coal, which pushed up the cost per MWh of nuclear power by 33%61.
We may observe that it is inconsistent to rely on the levelized cost method in an economy
with liberalized energy markets. The rationale used to establish the price of electricity, which
balances income and expenditure, including the remuneration of capital, is more in keeping
with regulated electricity tariffs set by the authorities. In a market economy, electricity
prices fluctuate; they are uncertain, just like the price of fuel consumed by generating
plants, or the price of tradable emissions permits. The solution is to use the conventional
method for calculating the return on a project in terms of net present value, while taking into
account the uncertainties. The price of electricity can thus be treated as a variable, which is
associated with a distribution function (e.g. a bell curve, on which the peak represents the
most probable expected value, and the extremities the lowest and highest values, of low
probability). Similarly various values with a range of probabilities are allocated to the other
variables affecting income or outgoings. Then we shake up all these data, carrying out
repeated random sampling, thousands of times – using the Monte Carlo method, in reference
to roulette. We thus obtain the risk profile for the investment, in other words a curve
showing the losses and gains it may produce, each level of loss and gain being associated
with a probability. If the curve is relatively flat the risk is high, because the probability is
more or less the same for low or high rates of return, both positive and negative. If the
curve rises sharply, the risk is low, with a substantial probability that the rate of return will
be centred near the peak, be it positive or negative. The merit of this probabilistic approach
is that it yields a mean value (obviously essential to know whether the return will be positive
or negative, low or high), but also an indication of the possible variances on either side of
the mean. Assisted by other authors62, Fabien Roques has used this approach to obtain a
better comparison of base load electricity generating technologies. With a whole series of
possible hypotheses – in particular a 10% discount rate – their research shows that gas
yields higher profits than nuclear power, at a lower risk, the latter point being due to the
gains achieved by more flexible plant operation. The load factor, instead of being constant
throughout the service life of plants, varies according to the market price of electricity. If the
price results in a loss, production stops, starting again when the net present value is once
again positive. A second interesting outcome of this work is that it puts figures on the
complementary relation between gas and electricity. A portfolio of assets, with gas-fired
plants making up 80% of capacity and nuclear power the remainder, yields a lower average
return than an exclusively gas-fired portfolio, but entails less risk. Investors may prefer this
combination which offers better protection, particularly from high, but unlikely losses
incurred if gas and carbon prices are high, a situation which has no effect on nuclear power.
60
See The Financing of Nuclear Power Plants, OECD Nuclear Energy Agency, 2009.
61
See footnote 18 [??] at the bottom of page 26.
62
Fuel mix diversification incentives in liberalized electricity markets: A Mean-Variance Portfolio theory
approach, Energy economics, 30, July 2008.
32/36
The competitive advantages of nuclear power and renewable energies
In suitable locations onshore wind farms display levelized costs comparable to those of
nuclear plants. Neither technology releases CO2 emissions and both are characterized by
high fixed costs. However, although nuclear power has a low marginal cost (about €6 per
MWh for fuel63), for wind the cost is zero. (The same is true of solar but, except under
extremely favourable conditions, its levelized cost is way above that of nuclear.) From an
economic point of view this difference is of fundamental importance, because in an electricity
market the optimal price is equal to the marginal cost of the marginal unit, in other words
the unit that needs to be generated to meet instantaneous demand. When instantaneous
demand is at its lowest point, generally in the middle of the night, only base load plants are
used (nuclear plants in France). If massive wind capacity were to be installed, the night
breeze would blow away gas and coal (perhaps even nuclear) during off-peak hours,
reducing their load factor and raising their respective costs per MWh. In fact the loss would
be even greater. Coal-fired or nuclear power stations do not ramp up to full capacity or shut
down instantaneously. So slowing down or stopping output at night would reduce the power
available in the early morning. To sell more electricity at times when prices are higher, it
may be in the interest of base load plant operators to bid negative prices in order to keep
their plants running all night. So at certain times of the day, large scale wind capacity would
result in a market price equal to its marginal cost, in other words zero, and even, at other
times, in a lower market price, equal to the opportunity cost of base load operators foregoing
a reduction in output.
Not taking into account variations in demand distorts the results when calculating the
levelized cost. Paul Joskow, at the MIT, has shown that this method is unsuitable for
intermittent renewables64. Only exceptionally are intermittent energies in synch with
demand. The wind does not blow harder at the beginning or end of the day, nor yet during
the five working days of the week, which is when power demand is highest. To simplify
matters we shall suppose that peak and off-peak hours are evenly distributed throughout the
year. We shall then suppose that an intermittent renewable plant produces two-thirds of its
output at off-peak hours, the remaining third at peak hours, and that its levelized cost per
MWh is the same as a base load plant. If the country as a whole needs one additional MWh
of power, the levelized cost method tells us that it makes no difference whether we invest in
wind power or a base load technology. Yet the second option is more useful because it will
produce proportionately more at peak hours: with all-year round output it operates half of
the time at peak hours, the other half off-peak. So the levelized cost method is biased
against investment in base load technology. It is also worth noting that it distorts the
ranking of intermittent renewable energies. As the sun does not shine at night, a solar plant
generally responds in a larger proportion to peak demand in summer than a wind farm. To
compare investment projects in various generating technologies, it is consequently wiser to
use the net present value method to estimate income on the basis of the hourly generation
profiles of plants and the electricity prices expected at different times of the day.
A third form of distortion which handicaps nuclear power is specific to Europe. The EU has
set targets for renewable energies. By 2020 renewables are slated to account for 20% of
final energy consumption. As applied to electricity this target means that renewables should
supply 35% of all electricity. Measures of this sort requiring a share of renewables in the
overall energy mix are commonplace. Most US states apply similar measures. But the EU is
unusual in that the measure operates in parallel with a carbon price. The EU system of
tradable emissions permits already adds to the cost of fossil-fuel generated electricity,
compared to nuclear power or renewables, changing their relative competitiveness in the
same way as a carbon tax. For example, with a permit costing €30 per tonne of CO2, it costs
63
Cour des Comptes report, p51.
64
Paul L. Joskow, Comparing the Costs of Intermittent and Dispatchable Electricity Generating
Technologies, American Economic Review, American Economic Association, vol. 101(3), May 2011,
p238-41.
33/36
about €30 per MWh more to generate 1 MWh using coal. Adding a target for renewables to
this scheme pushes the price of carbon down. The 20% target for 2020 was set without
adjusting the cap on CO2 emissions decided when the Emissions Trading Scheme was
originally set up. As a result the cut in emissions, made compulsory by the renewables
quota, restricts demand for permits. So their price drops. David Newbery has estimated that
the price of permits will be driven down by €10 per tonne by 2020, from €60 to €5065. To
avoid this downward pressure, the cap on emissions should have been lowered to allow for
the volume of CO2 recently avoided, in such a way as to achieve the target of 35% electricity
from renewable sources. In conclusion, the quota for renewable energies in the EU energy
mix has a dual effect. It deprives nuclear power of part of its potential market, despite it also
being carbon-free, and makes it less competitive by doing less to increase the price of
competing base load technologies, due to a lower carbon price.
We need to see the electricity system as a whole in order to grasp the relative competitive
advantages of nuclear power and renewables. As wind, solar and wave are intermittent
energy sources, and storing electricity is very expensive, large-scale development of
renewables involves building backup capacity to make up for the lack of wind, sunlight or
tide at certain times. Such capacity is far from negligible. For Ireland to meet its target for
the 2020 renewables quota, it will have to install 30 GW more renewable capacity, while
providing a further 15 to 20 GW of non-intermittent capacity as a backup66. To enable such
supplementary capacity to be built, the country must either agree to stupendous electricity
prices (several thousands of euros per MWh) at certain times of day, or set up capacity
markets to pay utilities even when they are not producing anything. Otherwise the plant will
simply not be built, because investors will anticipate difficulties covering fixed costs due to
the insufficient load factor. Nuclear power, dogged by higher fixed costs than gas and less
flexible production, is ill suited to catering for this new demand. All other things being equal,
the more intermittent energies develop, the more the competitiveness of nuclear power with
regard to gas will be undermined.
Looking beyond 2020 we see no sign of a possible improvement in the competitiveness of
nuclear power compared with renewables: quite the opposite. The development of storage
technologies and ongoing learning effects for wind and solar represent serious threats. Using
batteries to store electricity is still outrageously expensive. So far the only alternative
solution to have been developed is pumped-storage hydroelectricity. This involves using
electrical pumps to raise water from one reservoir to another at a higher elevation.
Meanwhile research is focusing on countless other possibilities. What results and applications
will research yield over the next 20 years? Without an answer it is hard to see whether
electricity storage will one day be sufficiently affordable to be deployed on a very large scale.
Realizing such a possibility would remedy the main shortcoming of intermittent renewable
energy and substantially increase its economic value, at least for renewables for which the
cost is currently close to that of more traditional technologies. This is the case for onshore
wind, setting aside the sources of distortion cited above. In the future it might also be the
case for offshore wind, and photovoltaic or concentrated solar. The costs of these
technologies have dropped substantially, with scope for powerful learning effects. But we
shall once again concentrate on terrestrial wind power. Its levelized cost per MWh was
divided by three, allowing for inflation, between the early 1980s and the late 2000s67.
Estimates indicate learning effects between 10% and 20%68. However a closer look reveals
that the reduction in the levelized cost in constant dollars stopped in 2005, and that the
levelized cost has actually risen since. Is this a sign that the technology has reached
65
Reforming Competitive Electricity Markets to Meet Environmental Targets, EPRG Working Paper, 1126,
Cambridge Working Paper in Economics, 1154
66
Quoted by Ambec and Crampes, 2010
67
See NREL 2012, IEA Wind Task 26, The Past and Future Cost of Wind Energy, WP2, May 2012
68
In a meta-study carried out for the International Panel on Climate Change, bearing on 18 estimates,
Wiser et al, 2011 (cited in the NREL study) suggest a 4% to 32% range. But the gap narrows to a 9% to
19% range, if only post-2004 estimates are considered.
34/36
maturity, with an end to diminishing costs? Very probably not, as shown by the report by the
US National Renewable Energy Laboratory. The rise in costs towards the end of the 2000s is
due to the increase in the price of materials and machinery, and a flattening out of
performance gains. But since then performance gains have started to improve and the cost
per MW of installed capacity is steady. The levelized cost across all wind speeds started
dropping again in 2012, down on 2009. The NREL has also done a comparison of 12
prospective studies looking ahead to 2030, covering 18 scenarios in all. Most of them predict
a 20% to 30% reduction in the levelized cost. Only one forecasts that it will remain steady.
These results obviously concern specific wind classes. The average performance of wind
capacity in a country or region may decline over time, due to the less favourable
characteristics of more recent locations, the first wind farms having occupied the spots with
the best conditions. The issue of siting is the only factor driving costs upwards. However in
the future it would be more than offset by the gains derived from mass production and
higher performance fed by R&D.
So nuclear has been caught in a pincer movement, so to speak. In OECD countries its high
cost, particularly regarding capital, is a handicap compared to gas. It is only competitive if a
carbon price is introduced. A fairly high one at that. In a drive to decarbonize its electricity,
playing on growth in renewables and replacement of its old nuclear power stations, the UK
has set a floor price for carbon in order to attract investors to its nuclear projects. Without a
carbon price, underpinned by a long-term commitment on its level, nuclear power no longer
makes the grade as a base load technology. On the other hand, setting aside onshore wind
power, it is still more cost-effective than intermittent renewables. So in principle there is
every reason why it should feature in a mix of carbon-free generating technologies. But only
in principle, because in practice it is sidelined and hampered by quotas for renewable
energies. In other countries nuclear power is at a disadvantage when compared to cheap,
polluting coal, but at least the prospects are a little better. Demand for energy is often so
great that all technologies are considered. Large countries such as China and India can
plausibly hope to reduce costs through large-scale production and learning effects. Smaller
nations may count on the advantage derived from keen competition between vendors of
turnkey solutions.
On reaching the end of the first part of this book, readers may feel slightly bereft, having lost
any sense of certainty regarding costs. There is no such thing as a ‘true’ cost for nuclear
power, which economists may discover after much trial and error. Nor yet are there any
hidden external costs, such as those related to managing waste or the risk of serious
accidents, which might completely change the picture if they were taken into account. Far
from reducing the cost of nuclear power, technical progress has actually contributed to its
increase. It makes no sense to assert that it is currently more or less expensive, in terms of
euros per MWh, to build a wind farm or a nuclear power station. There can be no universally
valid ranking order for coal, gas and the atom based on the cost of generating electricity.
But the loss of such illusions should not leave readers in a vacuum. The first part has also
provided a firm basis for assessing the costs of electricity, which depend on location and
various hypotheses on future developments. Consequently such costs can only be properly
calculated with a clear understanding of both factors. The construction cost of a nuclear
power station is not the same in Finland, China or the United States. Overall expenditure
may vary a great deal depending on the influence of the safety regulator, scale effects and
the cost of capital. Regarding wagers, the future prices of gas, coal and carbon dioxide will
be largely decisive in the ranking of coal, gas and nuclear power. These same prices will also
affect the profit margins of nuclear plants, their revenue depending on the number of hours
per year during which they operate, and whether the prices per kWh during those hours are
decided by a marginal generating plant burning coal or gas, or one powered by sunlight or
wind. Confronted by the risky long-term wagers which investors must make to calculate
costs and take decisions, even the most laissez-faire public authority will feel obliged to
intervene. Concerned by the general interest, it must set a discount rate, yet this is the
35/36
parameter with the greatest impact on the cost of nuclear power. This particular wager
hinges on how prosperous future generations may be: the richer they are, the lower the
discount rate will be, making nuclear power that much cheaper. Furthermore there is a
political choice to be made, in order to maintain a certain degree of equity between rich and
poor, and between generations, a choice which influences the rate set for converting present
euros into future euros.
What is more, analysing trends for past costs throws light on their future behaviour.
Historically nuclear technology has been characterized by rising costs. Today’s thirdgeneration reactors are no exception to this iron rule. They are safer than earlier
counterparts, but also more expensive. The escalation of costs may stop, but only on two
conditions: through a massive scale effect – if China chooses one type of reactor and sticks
to it, it may achieve this effect – or through a fundamental change in direction of innovation
– giving priority to modular design and small reactors, for instance. Failing this, nuclear
technology seems doomed to suffer a steady decline in its competitiveness compared with
any thermal technologies spared by taxes and renewable energies boosted by high learning
effects.
Setting aside any consideration of possible accidents, it would be an economically risky
choice for an operator to invest in building new nuclear power stations or for a State to
facilitate such projects. A daring bet indeed!
36/36